Spatial Dynamics and Ecology of Large Ungulate Populations in Tropical Forests of India [1st ed.] 9789811569333, 9789811569340

Citation preview

N. Samba Kumar  K. Ullas Karanth · James D. Nichols  Srinivas Vaidyanathan · Beth Gardner  Jagdish Krishnaswamy

Spatial Dynamics and Ecology of Large Ungulate Populations in Tropical Forests of India

Spatial Dynamics and Ecology of Large Ungulate Populations in Tropical Forests of India

N. Samba Kumar • K. Ullas Karanth •  James D. Nichols • Srinivas Vaidyanathan •  Beth Gardner • Jagdish Krishnaswamy

Spatial Dynamics and Ecology of Large Ungulate Populations in Tropical Forests of India

N. Samba Kumar Centre for Wildlife Studies Bengaluru, Karnataka, India

K. Ullas Karanth Centre for Wildlife Studies Bengaluru, Karnataka, India

James D. Nichols University of Florida Gainesville, FL, USA

Srinivas Vaidyanathan Foundation for Ecological Research, Advocacy and Learning Morattandi, Tamil Nadu, India

Beth Gardner School of Environmental and Forest Sciences University of Washington Seattle, WA, USA

Jagdish Krishnaswamy Centre for Biodiversity and Conservation Ashoka Trust for Research in Ecology and the Environment Bengaluru, Karnataka, India

ISBN 978-981-15-6933-3    ISBN 978-981-15-6934-0 (eBook) https://doi.org/10.1007/978-981-15-6934-0 © The Editor(s) (if applicable) and The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2021 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

It is impossible for us to convey in the dry pages of text here, the sheer scale of all the hard field work, high motivation, intense focus and passion for wildlife, brought to bear on the task of massive data collection by our army of volunteers, as well as dedicated staff and field assistants at the Centre for Wildlife Studies during 1989–2017. They mastered the art of bravely sneaking past wild elephants, dodging angry sloth bears, and even worse, literally being eaten alive by tiny jungle ticks that burrowed into their skins. These “transect surveyors” stayed in dingy field camps, ate monotonous meals and withstood the harsh work regimen and disciplinary code imposed on them. The high quality and massive quantity of the data we present here are testimonies to their commitment. We humbly dedicate this monograph to all these citizen scientists. It has been our privilege to work with them for over three decades.

Foreword

I met Samba Kumar in the summer of 2007 when he spent a sabbatical at USGS Wildlife Research Center at Patuxent to work on analyses related to his PhD effort, the effort that ultimately resulted in this monograph on ungulate ecology and conservation. At the time, one of the coauthors, Beth Gardner, was a post-doc in our group. To my mind, this time was Patuxent’s heyday not only in terms of staffing and scientific activity but also its vibrant and dynamic atmosphere. In those days, some of the hierarchical modeling ideas applied in this monograph were just being developed by members of our group, our collaborators, and colleagues. At least some of the ideas that came out of Patuxent during this period were inspired by collaboration with Ullas Karanth who approached me through Jim Nichols who has been a long-­term mentor to both of us. This collaboration between USGS Patuxent Lab and Centre for Wildlife Studies in India contributed to innovations such as the application of spatial capture–recapture models to camera trap data on tigers, occupancy models to sign survey data on mammals as well as the spatial distance sampling models to line transect data on ungulates, presented in this monograph. My engagement with spatial distance sampling work done by Samba Kumar was a part of this broader collaboration, which continues to this day (Karanth and Nichols 2017). This monograph by Samba and his coauthors is really two monographs rolled-up into one. On one hand, it is a monograph on spatial ecology and conservation of large ungulates in an important ecosystem, the Nagarahole–Bandipur landscape of southern India. This critical ecosystem includes a suite of ungulate species and also Indian tigers which several of the authors have been studying for most or all of their careers. Second, it is a monograph on applied hierarchical modeling. The important ecological and conservation context of this work is established in Chap. 1. As the authors emphasize, there is a sparsity of reliable information on ungulate abundance and distribution in south Asia, and this study addresses that information gap. The authors emphasize a number of important ecological and methodological challenges which structure the development and implementation of their study, statistical models, and analysis of the data. This sets the stage for the second viewpoint that this monograph is also a unique and very practical monograph on applied hierarchical modeling, and specifically what I would call “hierarchical distance sampling” (HDS). Distance sampling is one of the most important methodologies in wildlife monitoring, and the type of data analyzed by Samba and coauthors are routinely collected in many different systems throughout the world. vii

viii

Foreword

Hierarchical distance sampling concerns the situation where transects or points are surveyed using a distance sampling protocol, but the underlying abundance or density of each sample unit varies spatially according to some ecological process. In HDS models, understanding the structure and dynamics of how density varies among units is a fundamental objective. In Chap. 2, the core methodological chapter of the monograph, the authors provide a comprehensive specification of the HDS model, allowing for group size to affect detection probability, covariates that affect density and detection, spatial variation in the form of a flexible model of spatial dependence, and Bayesian variable selection. In Chap. 3, the HDS model is applied to the suite of five ungulate species using a number of covariates which represent habitat, physical environment, and management effectiveness. Together, these approaches allow geographic predictions of density for these important ungulates that occupy various niches in this community. One novel methodological aspect of the monograph is that the model is a novel integration of density surface models (DSM; Miller et  al. 2013) with a flexible model of spatial dependence (a conditional autoregression or CAR model). The authors provide a “fully Bayesian” implementation of the DSM framework in which both observation and process models are analyzed jointly, in contrast to the original two-stage analysis of Hedley and Buckland (1999). This monograph is the most comprehensive application of fully Bayesian DSMs in the literature, and it is one element which will be widely appreciated. In Chap. 3, the authors give a comprehensive analysis of an extraordinary monitoring data set representing more than 1400 km of survey transects, and in the appendix they provide a complete R script to show how the models are analyzed in NIMBLE which greatly increases the practical utility of this work. The management context of this study is detailed excellently in Chaps. 4 and 5. Chapter 4 is about quantifying the spatial distribution of proximate threats to ungulates, the development of a “composite human disturbance index” or CDI, and evaluating management strategies to increase ungulate abundance. The authors formulate various management strategies involving elements of habitat management and law enforcement and then project the impact of those actions on ungulate density under new habitat scenarios induced by six alternative management regimes. The authors look at existing strategies and budgets to compute a “return on management efforts.” Anti-poaching measures provide the most “bang for buck,” which suggest changes to the existing management regime. The final chapter summarizes the big picture of ungulate conservation in this system and, importantly, outlines a vision for the future derived from key conservation, monitoring, models, and management concepts developed as part of this work. The authors envision a formal adaptive management framework for the management of this landscape integrating decision-making with the biological (density/habitat models) and conservation context (increasing ungulate populations). This is perhaps an aspirational vision but very much consistent with the direction of the conservation field as a whole including the work of some of the monograph’s authors. In summary, “Spatial dynamics and ecology of large ungulate populations in tropical forests of India” represents a complete treatise on applied hierarchical

Foreword

ix

modeling motivated by an important conservation problem, the development of a rigorous methodological framework along with the provision of an accessible implementation, and then the use of the model to evaluate explicit management actions. As an important ecological case study, or as a methodological exposition, this monograph is sure to be widely useful and informative. I hope wildlife managers, researchers, and conservationists who are investing much energy and resources on recovering all threatened ungulate species will avidly consider recommendations provided here in their efforts. USGS Patuxent Wildlife Research Center Laurel, MD, USA March 27, 2020

J. Andrew Royle

References Hedley SL, Buckland ST, Borchers DL (1999) Spatial modelling from line transect data. J Cetacean Res Manage 1(3):255–264 Karanth KU, Nichols JD (eds) (2017) Methods for monitoring tiger and prey populations. Springer Nature Singapore Pte. Ltd. Miller DL, Burt ML, Rexstad EA, Thomas L (2013) Spatial models for distance sampling data: recent developments and future directions. Methods Ecol Evol 4(11):1001–1010

Preface

Wild ungulates inhabiting tropical forests of the world are bearing the brunt of increasing human impacts of the current Anthropocene epoch. Once ubiquitous and abundant throughout their range, distributions and abundances of most species are declining at unprecedented rates. In spite of the recognition of the critical role of wild ungulates in maintaining the structure, stability and functioning of ecosystems they live in, science-based efforts focused at ungulate conservation are scarce, particularly in the tropical regions where ungulates attain greatest diversity. Before the second half of the twentieth century, conservation efforts directed at populations of wild ungulates were constrained by the absence of rigorous methods to evaluate ungulate-habitat relationships to prioritize effective conservation interventions. Subsequent development of better field survey methods and robust statistical inferential approaches have laid the foundations for improved monitoring. However, the actual implementation of these improved methods by wildlife managers as well as researchers has consistently lagged in tropical forested regions, which are inherently more difficult environments for conducting monitoring compared to Savannas and other open habitats. The lack of adoption of modern monitoring, we believe, is also—at least partially—due to the apathy of wildlife professionals in Asia. Numerous reports of methods aimed at “censusing” ungulates (in other words getting a total count), which is an impossibility in reality, are examples of this weakness. This managerial apathy to rigorous science arises also because wildlife researchers have generally not published large-scale case studies illustrating useful application of such science. Several factors that compound challenges faced in monitoring ungulate populations need urgent attention. Most tropical forest ungulate species occur at relatively lower densities and consequently require field efforts to attain sufficient sample sizes demanded by design-based inference methods to estimate population density and abundance. The problems get further compounded because field counts of ungulates simply do not address measurement errors, most important being imperfect detection, leading to the “noise” in the data often drowning the “signal” sought by the surveyors. This situation characterizes virtually all approaches to abundance estimation, including the case of distance-sampling-based line transect survey, which traces its history to the middle of the twentieth century, and is widely used for ungulate

xi

xii

Preface

species that can be visually detected. Early exposure of some of the authors to these issues set the stage for the evolution of this monograph. In 1989, the first author (Kumar), then an amateur naturalist, participated as a volunteer in the line transect survey of ungulate species conducted by the second author (Karanth). The survey objective was to rigorously assess the prey availability for tigers and their sympatric predators in a macro-ecological study forming the latter’s graduate work at University of Florida. Karanth became deeply interested in line transect methodology, through his early interactions with field biologists Mel Sunquist, John Seidensticker, R. Rudran, and John Eisenberg. However, most crucially, he was also guided toward the right statistical approaches to the line transects by David Anderson at Colorado State University (in 1987) and James Nichols (third author) in 1989. In 1994, Kumar decided on a career in wildlife science, and joined as lead researcher in Karanth’s team at the Centre for Wildlife Studies, after quitting his position as a project manager in India’s space agency. Soon their focus progressed from implementing rigid field survey protocols (to ensure quality of data collected) to survey design, analysis, and estimation methods rooted in modern sampling theory applied under a likelihood-based inferential framework. The progress toward modern line transect methodology was enriched and facilitated greatly in 2000 by interactions with Stephen Buckland, David Borchers, Len Thomas, and Samantha Strindberg at the University of St. Andrews, which had by then become a global leader in this field. The continuing mentorship by the versatile James Nichols who straddled both the universes of likelihood theory and Bayesian inference, led Kumar and Karanth to the exciting field of hierarchical modeling of wildlife populations under a Bayesian paradigm pioneered by researchers at the US Geological Survey—William Link, Andy Royle, and Robert Dorazio. What impressed us most was the ability of their hierarchical models to elegantly deal with many thorny realities in the data generated from field sampling, without distracting us from ecological parameters that really mattered most. Along the way, other coauthors joined these efforts strengthening the team of authors. The most exciting aspect of working in such a team, consisting of passionate field biologists, quantitative ecologists, landscape ecologists, and mathematical statisticians, has been the ability we gained collectively to bridge the yawning gap between the rapidly unfolding methodological advancements and their full-fledged implementation in macro-ecological studies conducted in real world. In this monograph, we address the application of statistically robust hierarchical modeling approaches for understanding and conservation of tropical forest ungulate populations. Our aim is to demonstrate model development that can address thorny field sampling issues typically glossed over in many monitoring schemes, and, the application of these models to mechanistically explain spatial abundance patterns of five threatened ungulate species in a conservation priority landscape. Here, we expose readers to different important components of a science-based animal population monitoring program.

Preface

xiii

We first set out the practical context in which ecologically realistic models should be built. We show how key process parameters governing ungulate-habitat relationships can be identified. We then provide the details of a hierarchical model structure on the foundation of standard distance sampling line transect survey methods without major modifications to field practices already in use. Our focus is on overcoming common deficiencies of current methods used in surveys of tropical forest ungulates. We then apply our sampling, modeling, and estimation methods, successively testing them on field data on five ungulate species that vary greatly in body size, population densities, sociality (clustering tendencies), habitat preferences, and evasive responses to survey personnel. In contrast with traditional line transect surveys where the focus is primarily on estimating animal detection probabilities and overall animal densities, our Bayesian density surface models generate additionally refined information relevant to both ecological and management questions without significantly adding to field survey efforts or investments. In the final section, we take the readers through a structured evaluation process that can rigorously assess management effectiveness in population recovery programs directed at threatened ungulate species. We emphasize this monograph is not a treatise on the hierarchical modeling approaches in general, on which there is an excellent array of publications. It is also not a tract on general ecology of tropical forest ungulates or on the autecology of the exemplar ungulate species, which are only cited where relevant. Instead, we hope this monograph will address the needs of researchers, teachers, and conservation practitioners who are interested in applying cutting-edge monitoring methods for conserving threatened populations of ungulates. We expect the users (or teams of users) to collectively possess sufficient knowledge of basic statistics, species biology, and complex environments in which they monitor species of interest. To meet this goal, we explicitly illustrate in detail different aspects of application of Bayesian hierarchical spatial modeling approaches. These details cover, for our case, survey design, field sampling protocols, kinds of data generated, developing and comparing appropriate models, and estimation of population parameters using the most relevant models. The spatial models we use can predict species densities at different spatial scales, unlike most current methods that can only estimate crude ecological densities at wider scales, and sometimes only generate “indices” or “presence only data,” which recent studies show are poor surrogates for parameters that are useful for science or conservation. We have included detailed codes including model script, covariate data structure examples, frameworks for ecological modeling of ungulate-habitat relationships and modeling the anthropogenic impacts that are likely to influence these relationships. We believe our methods, with appropriate modifications, can also be applied for conservation threat assessments to meet their site-specific conservation needs. Based on these considerations, we believe this monograph offers methods and results that have relevance for addressing global challenges of conserving tropical forest ungulates in many regions beyond our wonderful field laboratory in

xiv

Preface

Nagarahole–Bandipur, which taught many things that simply cannot be learned within four walls of a conventional experimental laboratory. Bengaluru, Karnataka, India Bengaluru, Karnataka, India  Gainesville, FL, USA  Morattandi, Tamil Nadu, India  Seattle, WA, USA  Bengaluru, Karnataka, India 

N. Samba Kumar K. Ullas Karanth James D. Nichols Srinivas Vaidyanathan Beth Gardner Jagdish Krishnaswamy

Acknowledgments

We would like to thank Jeffrey Andrew Royle for his insightful comments and guidance in the process of developing the models that are at the core of this book. We also thank Andy for his generous and gracious foreword. Robert Dorazio subsequently spotted some weaknesses in the observation process part of the original model and helped us refine it further to address them. We are grateful for his guidance and support. We also thank James Hines for his contributions to the programming required in the early stages of model development. The field data used in this monograph are a part of a long-term project on predator–prey relationships at macro-ecological scales implemented by the Centre for Wildlife Studies, with funding support from the Wildlife Conservation Society, Liz Claiborne-Art Ortenberg Foundation, Ministry of Science and Technology, Government of India and several other donors. We are grateful to all these institutions. We acknowledge necessary permissions received from the Forest Department of Karnataka State and Ministry of Environment, Forest and Climate Change, Government of India. Individual Forestry officials and staff in Bandipur and Nagarahole are also thanked for their support. We are particularly thankful to the following senior forest officials: U. T. Alva, M. K. Appayya, A. C. Lakshmana, P. K. Sen, R. M. Ray, P. J. Dilip Kumar, B. K. Singh, Deepak Sarmah, and G. S. Prabhu who were instrumental in facilitating the overarching research project at various stages. We are indebted to institutions that supported individual authors involved in this publication. They include, respectively, Centre for Wildlife Studies (Samba Kumar), Wildlife Conservation Society, New York (Ullas Karanth), USGS Patuxent Wildlife Research Center (James Nichols), Foundation for Ecological Research, Advocacy and Learning (Srinivas Vaidyanathan), University of Washington (Beth Gardner), and Ashoka Trust for Research in Ecology & the Environment (Jagdish Krishnaswamy). We thank G. Vishwanatha Reddy, Indian Forest Service, for sharing his data on plant identification and human impacts in a part of the study area. We recognize him as one of those rare Indian wildlife managers who genuinely understood and valued the critical role of rigorous science in recovering wildlife. We are also grateful to Shridhar D. Bhat, Shrikant Gunaga, K. K. Sampath Kumara, and Karthik Teegalapalli for their help in plant identifications, Sanketh Shetty for assistance in compiling and validating the plant list and Chandan Pandey for overall computing support. xv

xvi

Acknowledgments

We thank Daniel Turek and Perry de Valpine of the NIMBLE Development Team who helped us to identify model initialization issues during the implementation of our hierarchical spatial model with indicator variables. We thank Aakanksha Tyagi and her team in Springer for their enthusiasm, patience, and superb assistance in the production and publication of this book. N. Samba Kumar K. Ullas Karanth James D. Nichols Srinivas Vaidyanathan Beth Gardner Jagdish Krishnaswamy Additional Acknowledgments from the Lead Author Finally, the lead author would like to acknowledge all his coauthors; it has been an immensely learning experience working with all of you. Ullas Karanth has been my mentor ever since he hired me to work for WCS in 1994, and a firm believer in my abilities. I have hugely benefitted from his clarity of thoughts and insights in the realm of conservation biology. Ever since I walked with Jim Nichols on a line transect in 1995, he has been my ideal. His simplicity, humility, and readiness to share his knowledge are phenomenal. He always found time to answer my questions and shaped my thinking about the way of conducting science. Andy Royle taught me the nuances of hierarchical modeling and showed incredible patience in guiding a hard core field biologist through a maze of complex models. My colleagues Arjun Gopalaswamy and Mohan Delampady patiently explained to me statistical concepts. Judd Howell, then Director of Patuxent Wildlife Research Centre, warmly facilitated my sabbatical in the summer of 2007. Beth Gardner taught me how to not get stuck in WinBUGS; Srinivas Vaidyanathan was my go-to-­ man every time I got stuck in issues such as GIS, remotely sensed data, data organization, and processing. Jagdish Krishnaswamy enthusiastically shared the eco-climatic distance data and co-supervised my PhD work. Over the years, I was fortunate to benefit from interactions with some capable forest officers: K. M. Chinnappa, S. N. Devaraju, A. T. Poovaiah, B. Venkatesh, D. Yatish Kumar, C. Srinivasan, C. K. Shivanna, and B. K. Singh foremost among them. My research team mates Raghavendra Mogaroy, Narendra Patil, Kiran Yadav, and Devcharan Jathanna, together with a dedicated field team, provided superb support both in and off the field. My long-term friends J.  Amarnath, G.  R. Sanath Kumar, D.  V. Girish, Krishna Narain (deceased), V.  Krishna Prasad, Praveen Bhargav, Darshan Khatau, and a host of other transect volunteers enthusiastically participated in transect surveys. I am grateful to all these individuals, as well as those whom I have not named here, for their support in my endeavors in many different ways. Last, but not least, I would have never switched my career from ISRO (Indian Space Research Organization) and succeeded as a wildlife biologist without the unflinching support from my wife Rohini, daughter Varsha and my parents. No words can adequately acknowledge their contributions. N. Samba Kumar

Contents

1 Introduction: The Conservation Issue ��������������������������������������������������    1 1.1 General Overview ����������������������������������������������������������������������������    2 1.2 Ungulates in South and Southeast Asia��������������������������������������������    3 1.3 Wild Ungulates in India��������������������������������������������������������������������    3 1.4 Environmental Drivers of Ungulate Distribution and Abundance����������������������������������������������������������������������������������������    5 1.4.1 Challenges of Ungulate Monitoring at Varying Spatial Scales����������������������������������������������������������������������    5 1.4.2 Key Factors Influencing Ungulate Distribution and Abundance��������������������������������������������������������������������������    6 1.4.3 Methodological Challenges������������������������������������������������    7 1.5 Study Goal and Specific Objectives��������������������������������������������������    9 1.6 Study Region, Landscape and Habitats��������������������������������������������   10 1.6.1 Nagarahole National Park ��������������������������������������������������   12 1.6.2 Bandipur National Park������������������������������������������������������   15 1.7 Study Species������������������������������������������������������������������������������������   17 1.7.1 Wild Pig (Sus scrofa)����������������������������������������������������������   17 1.7.2 Red Muntjac (Muntiacus muntjak) ������������������������������������   18 1.7.3 Chital (Axis axis)����������������������������������������������������������������   19 1.7.4 Sambar (Rusa unicolor)������������������������������������������������������   20 1.7.5 Gaur (Bos frontalis)������������������������������������������������������������   21 1.8 Ecological Comparisons ������������������������������������������������������������������   23 1.9 Organization of the Monograph��������������������������������������������������������   23 References��������������������������������������������������������������������������������������������������   25 2 Development of Hierarchical Spatial Models for Assessing Ungulate Abundance and Habitat Relationships����������������������������������   35 2.1 Introduction��������������������������������������������������������������������������������������   36 2.2 Modeling Philosophy������������������������������������������������������������������������   38 2.2.1 Overview of Distance Sampling and Line Transect Sampling ����������������������������������������������������������������������������   38 2.2.2 Model-Based Inferential Approaches ��������������������������������   39 2.2.3 Hierarchical Models������������������������������������������������������������   40

xvii

xviii

Contents

2.3 Model Development��������������������������������������������������������������������������   40 2.3.1 Motives and Considerations������������������������������������������������   40 2.3.2 Model Formulation ������������������������������������������������������������   41 2.3.3 Model Notations�����������������������������������������������������������������   41 2.3.4 Observation Model��������������������������������������������������������������   42 2.3.5 Abundance (Process) Model ����������������������������������������������   46 2.3.6 Modeling Effects of Site-Level Covariates������������������������   46 2.3.7 Modeling Effects of Landscape-Level Covariates��������������   47 2.3.8 Modeling Spatial Effects����������������������������������������������������   47 2.3.9 Modeling the Variation in Sampling Efforts ����������������������   49 2.3.10 Modeling the Effects of Spatial Misalignment ������������������   49 2.4 Bayesian Inference����������������������������������������������������������������������������   50 2.5 Bayesian Variable Selection��������������������������������������������������������������   50 2.6 Example of Application of the Hierarchical Spatial Model��������������   52 2.6.1 Line Transect Sampling Data on Chital Populations����������   52 2.7 Results����������������������������������������������������������������������������������������������   54 2.7.1 Detection Function��������������������������������������������������������������   54 2.7.2 Chital Abundance����������������������������������������������������������������   54 2.7.3 Effects of Predictor Variables on Abundance����������������������   56 2.7.4 Relative Importance of Predictor Variables������������������������   57 2.8 Discussion ����������������������������������������������������������������������������������������   57 Appendix 2.1����������������������������������������������������������������������������������������������   61 Example Data Structure of the Four Input Files Used for the Bayesian Hierarchical Spatial Modeling������������������������������������   61 Appendix 2.2����������������������������������������������������������������������������������������������   64 R Script for Fitting the Bayesian Hierarchical Spatial Model ��������������   64 References��������������������������������������������������������������������������������������������������   78 3 Model-Based Assessment of Ungulate-­Habitat Relationships ������������   83 3.1 Introduction��������������������������������������������������������������������������������������   84 3.2 Field Survey Data on Ungulate Species��������������������������������������������   86 3.3 Predictors of Ungulate Abundance ��������������������������������������������������   87 3.3.1 Identification of Ecological and Management Covariates����������������������������������������������������������������������������   87 3.3.2 Covariates Influencing Ungulate Abundance����������������������   88 3.4 Data Analyses������������������������������������������������������������������������������������   91 3.5 Results����������������������������������������������������������������������������������������������   92 3.5.1 Encounter Rate, Detectability and Cluster Size of Ungulate Species����������������������������������������������������������������   92 3.5.2 Density and Abundance of Ungulate Species ��������������������   95 3.5.3 Predictors of Ungulate Species Abundance������������������������  100 3.5.4 Relative Importance of Predictors of Abundance ��������������  101 3.6 Discussion ����������������������������������������������������������������������������������������  104 3.6.1 Ungulate Abundance: Implications for Management ��������  104 3.6.2 Predictors of Ungulate Abundance ������������������������������������  105 3.6.3 Relative Importance of Predictors of Abundance ��������������  107

Contents

xix

Appendix 3.1����������������������������������������������������������������������������������������������  108 List of Plants Encountered in Vegetation Plots Laid Along Transects in the Nagarahole-Bandipur Study Landscape��������������������������������������  108 Appendix 3.2����������������������������������������������������������������������������������������������  121 Description of Human Activity Signs Encountered Along Line Transects During the Human Impact Survey in the NagaraholeBandipur Study Landscape and Their Metrics of Field-Measure����������  121 Appendix 3.3����������������������������������������������������������������������������������������������  123 Summary of the Posterior Distributions of Detection and Abundance Parameters Together with Their Monte Carlo Standard Errors (MCse) and Posterior Density Plots for Each of the Study Species in the Nagarahole-­Bandipur Landscape, India������������������������������������������������  123 Wild Pig ������������������������������������������������������������������������������������������������  123 Muntjac��������������������������������������������������������������������������������������������������  130 Chital������������������������������������������������������������������������������������������������������  134 Sambar ��������������������������������������������������������������������������������������������������  145 Gaur��������������������������������������������������������������������������������������������������������  151 References��������������������������������������������������������������������������������������������������  162 4 Assessing Threats to Ungulates and Management Responses ������������  167 4.1 Introduction��������������������������������������������������������������������������������������  168 4.2 Methods��������������������������������������������������������������������������������������������  169 4.2.1 Spatial Distribution of Proximate Threats��������������������������  169 4.2.2 Assessment of Management Actions����������������������������������  174 4.3 Results����������������������������������������������������������������������������������������������  176 4.3.1 Spatial Distribution of Proximate Threats��������������������������  176 4.3.2 Assessment of Management Actions����������������������������������  177 4.4 Discussion ����������������������������������������������������������������������������������������  179 4.4.1 Assessment of Threats��������������������������������������������������������  179 4.4.2 Assessment of Management Actions����������������������������������  179 References��������������������������������������������������������������������������������������������������  182 5 Conservation of Tropical Forest Ungulates: The Way Forward����������  185 5.1 The Conservation Context����������������������������������������������������������������  186 5.2 The Conservation Issue��������������������������������������������������������������������  187 5.3 Utility of Hierarchical Models as a General Methodological Tool ������������������������������������������������������������������������  188 5.4 Summary Review of Findings����������������������������������������������������������  189 5.5 Conservation Implications����������������������������������������������������������������  189 5.6 Conservation of Tropical Forest Ungulates: The Way Forward��������  190 References��������������������������������������������������������������������������������������������������  192

About the Authors

N. Samba Kumar   is currently an Emeritus Scientist at the Centre for Wildlife Studies following his retirement as Director of Research and Training. He holds a master’s degree in Ecology from Pondicherry University and a PhD from Manipal University, India. His research focuses on disentangling complex factors that shape abundance patterns of large mammals and on developing rigorous field survey and analytical methods for monitoring large mammal populations at multiple spatial scales. He has authored/coauthored over 75 scientific articles/reports and one book. K. Ullas Karanth   is the Director of Centre for Wildlife Studies, India, and a Fellow of the Indian Academy of Sciences. He graduated in Wildlife Biology from the University of Florida, Gainesville, USA, and Mangalore University, India, and served as the Director of the Wildlife Conservation Society-India program (1988–2017). His research focuses on carnivore ecology and wildlife management. He has authored over 150 scientific articles and 15 technical and popular books. James D. Nichols   recently retired from his position as a government scientist with the U.S. Fish and Wildlife Service, the U.S. National Biological Survey, and the U.S.  Geological Survey. After studying at Louisiana State University and Michigan State University, his career has focused on animal population dynamics, biostatistics, and decision-making in conservation. He has authored over 400 scientific articles and 10 books. Srinivas  Vaidyanathan   is a Senior Research Fellow and trustee of the Foundation for Ecological Research, Advocacy and Learning (FERAL), India. He graduated in Ecology from Pondicherry University, India. His research focuses on landscape ecology, space–time analysis, and connectivity conservation, with a particular interest in understanding the changes in landscape-level structure and processes that affect the distribution and abundance of large mammal populations. He has authored/coauthored over 50 scientific articles/reports and 4 books. Beth  Gardner   is an Associate Professor at the University of Washington’s School of Environmental and Forest Sciences, in Seattle. She holds MS and PhD degrees from Cornell University. She has authored/coauthored over 70 scientific

xxi

xxii

About the Authors

articles and one book on capture–recapture models. Her research focuses on statistical models, ecology, and population dynamics. Jagdish Krishnaswamy   is a Senior Fellow at the Ashoka Trust for Research in Ecology and the Environment (ATREE), Bengaluru, India. He holds an M.S. degree in Statistics and a PhD in Environmental Science from Duke University, NC, USA. His main areas of research and teaching are ecohydrology, landscape ecology, and applied statistics, and he has published over 60 peer-reviewed articles and several book chapters.

List of Figures

Fig. 1.1 Map of the study landscape showing Nagarahole and Bandipur National Parks and adjacent forests��������������������������������������������������  10 Fig. 1.2 Land-use patterns around the study landscape of Nagarahole-Bandipur and the resulting indicative patterns of anthropogenic pressures on the study area����������������������������������������   12 Fig. 1.3 Map showing human settlements and official protection patrol camps in and around the study landscape of NagaraholeBandipur��������������������������������������������������������������������������������������������  13 Fig. 2.1 Estimated relationship between chital cluster size group specific detection probabilities and the perpendicular distance from the transect line for four categories of cluster size groups in the Nagarahole-Bandipur landscape in India. The numbers in the legend represent the expected number of individual animals in each cluster size group, and the shaded regions indicate 95% credible limits of the detection probability of respective cluster size group������������������������������������������������������������������������������  54 Fig. 2.2 Spatial distribution of chital density (estimated posterior median) at fine scale resolution (1-km2) showing ‘hot-spots’ of local abundance in the Nagarahole-Bandipur study landscape in India��������  56 Fig. 3.1 Map showing the system of line transects in the NagaraholeBandipur study landscape ����������������������������������������������������������������  86 Fig. 3.2 Estimated relationship between the probability of cluster size detection and the perpendicular distance from the transect line for each of the cluster size categories of five ungulate species in the Nagarahole-Bandipur landscape in India (a) Wild pig, (b) Muntjac, (c) Chital, (d) Sambar, (e) Gaur. The numbers in the legend represent expected number of individual animals in each cluster size category��������������������������������������������������������������������������  93 Fig. 3.3 Spatial distribution of the estimated fine-scale (1-km2 grid-cell level) density of animals (posterior median; number of individuals/km2) of five ungulate species and their abundance ‘hot spots’ in the Nagarahole-Bandipur study landscape. (a) Wild pig, (b) Muntjac, (c) Chital, (d) Sambar, (e) Gaur ������������  97 xxiii

xxiv

List of Figures

Fig. 4.1 Map of the Nagarahole-Bandipur study landscape showing village settlements��������������������������������������������������������������������������������������� 170 Fig. 4.2 Land-use and cropping patterns around the Nagarahole-Bandipur study landscape and the resulting patterns of anthropogenic pressures on the study landscape������������������������� 170 Fig. 4.3 Map showing forest protection (patrol) camps in and around the Nagarahole-Bandipur study landscape����������������������������������������  171 Fig. 4.4 Spatial distribution of human impacts interpolated from the regressor effects of distance to human settlements and distance to protection (patrol) camps in the Nagarahole-Bandipur study landscape������������������������������������������������������������������������������������������  177 Fig. 4.5 Distribution of natural and artificial water tanks, perennial streams and reservoirs in the Nagarahole-Bandipur study landscape������������������������������������������������������������������������������������������  180

List of Tables

Table 2.1 Two-dimensional structure of example line transect count data grouped in distance categories��������������������������������������������������������  43 Table 2.2 Three-dimensional structure of example line transect count data grouped in cluster size categories and distance categories ����������������������������������������������������������������������������������������   44 Table 2.3 Estimated posterior mean, median and 95% credible limits of cluster-size group specific detection probabilities (gs) of chital in the Nagarahole-Bandipur study landscape����������������������������������   55 Table 2.4 Estimated posterior mean, median and 95% credible limits of group size mean (grszMean) of each cluster size category of chital in the Nagarahole-Bandipur study landscape������������������������   55 Table 2.5 Estimated posterior mean, median and 95% credible intervals for the effects of covariates on expected cluster abundance of chital in the Nagarahole-Bandipur landscape����������������������������������   57 Table 2.6 Estimated posterior mean weights and their associated Monte Carlo standard errors (MCse) of indicator variables for each covariate effect included in the chital abundance model ����������������   57 Table 3.1 Predicted response of five study species to each of the abundance predictors in the Nagarahole-Bandipur landscape in India based on a priori hypotheses����������������������������������������������   88 Table 3.2 Estimated posterior mean and 95% Highest Posterior Density (HPD) credible intervals of the effective half-strip width (in meters) for detections of five ungulate species in the Nagarahole-­Bandipur study landscape��������������������������������������������   96 Table 3.3 Estimated posterior mean and 95% Highest Posterior Density (HPD) credible intervals of the cluster-size (expected number of individuals in any observed cluster) for five ungulate species in the Nagarahole-Bandipur study landscape����������������������������������   96 Table 3.4 Summary of the spatial distribution of the estimated posterior medians of fine-scale (1-km2 grid-cell level) population densities of ungulate clusters (number of clusters/km2) and individuals (number of individuals/km2) of five study species within the sampled area of Nagarahole-­Bandipur landscape ��������������������������   99 xxv

xxvi

List of Tables

Table 3.5 Estimated overall density (number of individual animals/km2) of five ungulate species within the sampled area of Nagarahole and Bandipur reserves��������������������������������������������������  100 Table 3.6 Estimated posterior mean, standard deviation (SD) and Monte Carlo standard error (MCse) together with 95% Highest Posterior Density (HPD) credible intervals for the effects of explanatory variables on expected cluster abundance of five ungulate species in the Nagarahole-Bandipur study landscape ����������������������������������������������������������������������������������������  101 Table 3.7 Estimated posterior means of indicator variables for each covariate effect included in the abundance models of five ungulate species in the Nagarahole-Bandipur study landscape������  102 Table 3.8 Ranking of explanatory variables based on their estimated posterior inclusion probabilities in the abundance models of five ungulate species in the Nagarahole-Bandipur study landscape ����������������������������������������������������������������������������������������  102 Table 3.9 Marginal posterior probabilities of the optimal set of models explaining spatial abundance variation of five ungulate species in the Nagarahole-Bandipur study landscape����������������������������������  103 Table 4.1 Description of human activities affecting ungulate habitat quality in the Nagarahole-­Bandipur landscape together with metrics of field-measure ����������������������������������������������������������������������������������  173 Table 4.2 Description of management actions implemented in the Nagarahole-Bandipur landscape to enhance quality of tiger-ungulate prey habitat along with their percent resource allocations over a 5-year period������������������������������������������������������  175 Table 4.3 Predicted range of percent changes in the abundance of five ungulate species in the Nagarahole-Bandipur study landscape in response to three levels of hypothesized increase in efforts under each management action category����������������������������������������  178 Table 4.4 Predicted range of percent changes in the abundance of five ungulate species in the Nagarahole-Bandipur study landscape in response to three levels of hypothesized decline in efforts under each management action category����������������������������������������  178

List of Appendices

Appendix 2.1 Example Data Structure of the Four Input Files Used for the Bayesian Hierarchical Spatial Modeling ��������������������������������   61 Appendix 2.2 R Script for Fitting the Bayesian Hierarchical Spatial Model����������������������������������������������������������������������������   64 Appendix 3.1 List of Plants Encountered in Vegetation Plots Laid Along Transects in the Nagarahole-Bandipur Study Landscape��������  108 Appendix 3.2 Description of Human Activity Signs Encountered Along Line Transects During the Human Impact Survey in the Nagarahole-Bandipur Study Landscape and Their Metrics of Field-Measure ����������������������������������������������  121 Appendix 3.3 Summary of the Posterior Distributions of Detection and Abundance Parameters Together with Their Monte Carlo Standard Errors (MCse) and Posterior Density Plots for Each of the Study Species in  the Nagarahole-Bandipur Landscape, India����������������������������  123

xxvii

1

Introduction: The Conservation Issue

Abstract

1. Large ungulates are major ecological drivers shaping the structure and functioning of terrestrial ecosystems. However, they are also a highly vulnerable group of mammals because of increasing human impacts in the form of hunting, habitat loss and degradation. 2. In view of the general decline of wild ungulates worldwide, and more specifically in tropical forests, there is an urgent need for rigorous assessments of population statuses of tropical forest ungulate species, as well as evaluations of drivers of their declines in order to make timely and informed conservation decisions. 3. In this monograph, we provide an overarching modeling framework and develop a set of specific methods required for rigorous analyses of ungulate populations. We also demonstrate their practical application by investigating spatial variation in ungulate abundance patterns and their key determinants in the case of five large, sympatric tropical forest ungulates in southwestern India. 4. This chapter provides the conservation context in which such understanding of ungulate-habitat relationships is necessary for the conduct of informed management. We also elaborate on specific environmental, logistical and statistical challenges involved in our macro-ecological field investigation. 5. The detailed prior information on the study landscape and a synthesis of current knowledge on biology and conservation issues affecting the study species are used for the formulation of a priori hypotheses about the drivers of wild ungulate abundance patterns in tropical forest systems. These hypotheses are tested by appropriately designed confrontations of plausible models against survey data in the subsequent chapters of this monograph.

© The Editor(s) (if applicable) and The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2021 N. S. Kumar et al., Spatial Dynamics and Ecology of Large Ungulate Populations in Tropical Forests of India, https://doi.org/10.1007/978-981-15-6934-0_1

1

2

1.1

1  Introduction: The Conservation Issue

General Overview

Ungulates are hoofed mammals adapted to a terrestrial, herbivorous life with fast locomotion (Macdonald 2001). They are distributed worldwide (except Australia and Antarctica) and exhibit a remarkable diversity of body sizes ranging from the 1-kg chevrotains to the 3-tonne hippopotamuses. Their diet is varied, including leaves, grasses, flowers, fruits and seeds of trees, herbs, roots, tubers and even insects and flesh. Consequently, their habitats occur in a wide variety of biomes ranging from desert, grassland, tundra to forests of various types. Living ungulate species are represented by 13 families, 95 genera and 257 species worldwide (Wilson and Reeder 2005). They constitute about 80% of all the terrestrial mammalian species with body mass of over 50 kg (Macdonald 2001). Ungulates are among the major ecological drivers shaping the structure and functioning of terrestrial ecosystems (Duncan et  al. 2006; Terborgh et  al. 2008). Studies show ungulates are “potential initiators” of interaction chains (Pringle et al. 2007), significantly influencing ecosystem processes (Augustine and McNaughton 2006). They impact plant community structure and forest dynamics directly through grazing and browsing (Chase et al. 2000; Adler et al. 2001), as well as indirectly through seed dispersal and seedling establishment (Paine 2000; Duncan et al. 2006; Prasad et  al. 2006). Ungulates also influence primary plant productivity, nutrient cycling and soil fertility both positively (McNaughton 1985; McNaughton et  al. 1997; Frank et al. 1998, 2002) and negatively (Archer 1995; Anderson and Briske 1995; Pastor and Cohen 1997; Ritchie et al. 1998). They also significantly influence soil processes (Bardgett and Wardle 2003). As an important constituent of trophic chains, ungulates are principal prey of large carnivores (Schaller 1967, 1972; Eisenberg 1980; Johnsingh 1983; Chellam 1993; Karanth and Sunquist 1995; Macdonald 2001) thereby shaping the structures of predator guilds (Sunquist et al. 1999; Karanth et al. 2004). Large ungulate species are declining worldwide today (Robinson and Bennett 2000; Owen-Smith and Mills 2006; Schipper et  al. 2008) becoming some of the most threatened mammals (Baillie et al. 2004; Ripple et al. 2015). Ungulates are inherently vulnerable for at least three sets of reasons. First, they are rare and demographically vulnerable because of traits such as large body size, substantial dietary and energetic needs, small litter size, and, long inter-calving interval (Eisenberg 1980; Hudson 1984; Madhusudan and Mishra 2003). Second, they are priority targets for hunting by humans for a variety of reasons (Robinson and Bennett 2000). Third, their habitats and forage resources are increasingly under pressure from expanding agriculture and animal husbandry. Not surprisingly, overhunting (Baillie et al. 2004; Corlett 2007), rapid habitat loss (Macdonald 2001) and habitat degradation due to overexploitation (Groom 2006) have driven sharp declines of ungulate species worldwide. In this context, assessment of status of ungulate populations and factors that influence it are of critical interest to wildlife managers and researchers. Estimating animal population responses to conservation interventions is a fundamental goal of all monitoring programs (Nichols and Williams 2006). As a “system state variable”, population size or density is also of interest to wildlife biologists, because it is the central component for investigating animal population dynamics (Williams et al. 2002).

1.3  Wild Ungulates in India

1.2

3

Ungulates in South and Southeast Asia

Although the ungulate species richness in Africa is the highest among all continents (94 species; Wilson and Reeder 2005), 15 countries in the South and Southeast Asian region also support 83 species of ungulates (Ahrestani and Sankaran 2016). The tropical Asian region occupies only half the area of the Neotropics, and only onethird the area of the Afrotropics, yet it supports three times more species than the former and an almost equal number of species as the latter region (Bibi and Metais 2016) attesting to its superior diversity and species richness. Within tropical Asia, Southeast Asia has a large number of endemic species because of the island speciation effect, whereas South Asia still supports many protected ungulate populations that are comparable to those in Africa in terms of population density and biomass (e.g., Karanth and Sunquist 1992). Furthermore, the above mentioned distribution and abundance patterns of Asian ungulates representatively capture the equally diverse and rich environments they inhabit thus serving as conservation flagships in many cases. Additionally, tropical Asia is characterized by rising human populations and their developmental aspirations. Consequently, these are exerting enormous pressures on ungulates and their habitats. However, the losses of ungulate species suffered in the tropical Asia region are not as extreme as in the temperate Eurasia or even the Americas (Bibi and Metais 2016). Their survival and resilience have been attributed to unique religious, cultural, and historical factors that have fostered a higher level of tolerance for ungulates in many sub-cultures (Ahrestani and Sankaran 2016). Overall, gaining an understanding of patterns of ungulate biodiversity and how these are shaped in this region is critically important for their conservation. Ungulates in tropical Asia are under severe threat due to overhunting, habitat loss and degradation. In the immediate term, these are all primarily anthropogenic rather than environmental factors. A closer examination of the assessment of the conservation status of tropical ungulates by IUCN (2019) shows that ~60% of these are in potential danger of extinction being classified as vulnerable, endangered or critically endangered (Ahrestani and Sankaran 2016). An additional 10% of the species are classified as data-deficient. Insufficient evidence or lack of quantitative information has precluded classification of ~27% of the species into categories of higher-­ extinction risk. Furthermore, the populations of all the ungulates species in tropical Asia have shown declines over the years (IUCN 2019). Therefore, a rigorous evaluation of the current population status of key threatened tropical forest ungulate species and drivers of their declines is urgently required to make informed conservation and management decisions (Sankaran and Ahrestani 2016). The monograph is a step in this direction.

1.3

Wild Ungulates in India

India is among the 12 global mega-biodiversity nations (Mittermeier and Mittermeier 2005), drawing its faunal elements from Indo-Malayan, Afro-tropical and Paleo-­ Arctic bio-geographic realms in addition to its endemic species (Briggs 2003). Diversity of larger ungulates is particularly rich in the Indian sub-continent, with 39 species nested in 23 genera, seven families and two orders (Prater 1985; Wilson and

4

1  Introduction: The Conservation Issue

Reeder 2005). These constitute ~15% of the extant ungulate species globally (Wilson and Reeder 2005). Ungulate population densities and biomasses in protected sites in the Indian subcontinent are comparable to some of the highest in Africa as suggested by limited number of studies available (Eisenberg and Seidensticker 1976; Karanth and Sunquist 1992; Khan et al. 1996; Kumar 2000). Moreover, the community structure, habitat relationships and habitat use patterns of ungulates in the Indian subcontinent have not been well-explored as they have been in quantitative studies of African ungulates (Peters and Raelson 1984; Du Toit and Owen-Smith 1989). Only some preliminary studies by Schaller (1967), Eisenberg and Lockhart (1972), and, Berwick (1974) were available until the 1970s. Thereafter, there have been a few more studies of population biology of ungulates in South Asia (Seidensticker 1976; Eisenberg and Seidensticker 1976; Martin 1977; Dinerstein 1979, 1980, 2003; Mishra 1982; Tamang 1982; Johnsingh 1983; Karanth and Sunquist 1992; Khan et al. 1996; Kumar 2000; Jathanna et al. 2003; Bagchi et al. 2004; Karanth et al. 2008; Ahrestani 2009; Wegge et al. 2009; Steinmetz et al. 2010; Wang 2010; Harihar et al. 2014). These studies have reported ungulate densities and biomass estimates from different parts of south Asia. As a part of its National Tiger Estimation (NTE) exercises, the National Tiger Conservation Authority (NTCA) in collaboration with the Wildlife Institute of India (WII) has tried to assess the abundance of ungulates across several protected reserves in India (Jhala et al. 2015). However, many of these reported estimates are unreliable because they are derived from ad hoc methods that lack application of rigorous sampling-based quantitative approaches developed in the last two decades (Williams et  al. 2002; Royle and Dorazio 2008; Link and Barker 2010). Furthermore, only a few studies have examined the patterns of ungulate abundance in relation to different ecological conditions (Schaller 1967, in Sal forests; Dinerstein 1979 and Mishra 1982, in swampy grasslands; Karanth and Sunquist 1992, in mixed deciduous forests; Khan et al. 1996, in arid forests; Bagchi et al. 2004, in semi-arid forests; Wang 2010, in temperate forests; see Puri et al. 1983; Meher-Homji 1990 for details of these forest types). Ungulates in India appear to attain high densities in well-protected moist-­ deciduous forests and alluvial grasslands (Karanth et al. 2004). However, their densities are massively depressed by human overhunting (Madhusudan and Karanth 2002; Karanth et al. 2004) and intense resource competition with livestock (Khan et al. 1996; Madhusudan 2004). Consequently, ungulate distribution and abundance have drastically declined across large parts of the Indian subcontinent (Karanth et  al. 2010). Specifically over the last 100  years, ungulates have become locally extinct at 52% of sites examined (range: 25–81%) compared to their historic distribution (Karanth et al. 2010). India’s forest cover has been lost to an extent of 80% in the past 300 years (Sanderson et al. 2006) with remaining habitat being shared with human and livestock populations. Relatively un-impacted ungulate habitats are now confined mostly to a network of protected nature reserves that constitute less than 5% of the land area (Karanth et al. 2009, 2010). An estimated 10 million people and 12 million livestock live either within or along fringes of India’s nature reserves (Directorate of Census Operations 2004) competing with wild ungulates for space

1.4  Environmental Drivers of Ungulate Distribution and Abundance

5

and food. This has led to extensive loss, degradation and fragmentation of ungulate habitats even in and around remaining nature reserves, in spite of major conservation efforts. However, successful conservation efforts do show that some ungulate populations can be recovered relatively rapidly through strong protectionist interventions (Karanth et  al. 1999) as well as management actions to enrich habitats (Karanth and Sunquist 1992; Khan et al. 1996; Karanth et al. 2008; Harihar et al. 2009). These factors set the context for our present study.

1.4

 nvironmental Drivers of Ungulate Distribution E and Abundance

We acknowledge that for recovering ungulate populations, clear understanding of the complex interplay of ecological forces, such as species-specific life-history traits, primary habitat productivity and environmental characteristics and anthropogenic pressures, is essential (Baillie et al. 2004). Obtaining reliable assessments of ecological determinants of ungulate abundance, however, poses many challenges. In the following sections of this chapter, we examine these challenges.

1.4.1 C  hallenges of Ungulate Monitoring at Varying Spatial Scales For the purpose of this monograph, we use the word “abundance” as a synonym for the state variable defined as ‘population size’ (Williams et al. 2002). Several ecological mechanisms influence processes that set ungulate abundance levels. Different species respond differently to environmental gradients including natural (e.g., temperature) and anthropogenic (e.g., livestock presence, hunting pressure) factors. Assessment of these processes at different spatial scales is a fundamental ecological challenge faced in monitoring. Reliable models of ungulate species abundance across space and time should be able to (1) estimate their local population densities, (2) estimate total abundances for a given study site at a given time, (3) assess population dynamic trends over time at the same site, and (4) identify key factors that influence both spatial and temporal patterns of these population dynamics. Body size and diet are the overarching intrinsic biological species traits that set limits on its abundance (Eisenberg 1980; Hudson 1984). Furthermore, the quality and quantity of external biotic and abiotic resources (e.g., food, nutrients, water, shelter, escape cover, terrain, soil, weather) available at different spatial scales will also influence distribution and abundance patterns of ungulates (Skidmore and Ferwerda 2008). Additionally, ungulates exhibit large variations in abundance owing to ecological factors that operate at different spatial scales. Although the concept of scale is loosely defined in ecology (Skidmore and Ferwerda 2008), it is important to understand animal responses to resources available at varying spatial scales. Generally, a spatial scale is defined based on studying species at two levels (Levin 1992): broad-scale or landscape level (regions of

6

1  Introduction: The Conservation Issue

>100–500 km2 area) and fine-scale or local level (sites of >1 km2 to 1500 landless families living inside and 5000 at the fringes; Appayya 2001), and the population consists of mainly Jenu Kuruba, Betta Kuruba and Yarava ethnic communities, who were originally hunter-gatherers but subsequently shifted to farm labor and subsistence farming (Karanth et al. 1999). Over the years, many of these landless families have voluntarily opted to relocate from park interiors to outside where they have better opportunities and infrastructure for integrating with the mainstream society.

1.6  Study Region, Landscape and Habitats

15

Nearly 50% of these families moved out under incentive-driven voluntary resettlement schemes, offered by the Government of India and implemented with critical support from local non-governmental organizations (Karanth et al. 2001; Karanth and Karanth 2007). The rest are involved in forest resource extraction and other allied activities, mainly feeding the market supply chain. Besides these impacts from within, the park is also subjected to a range of anthropogenic pressures resulting from varied land-use patterns at its fringes, illustrated in Fig. 1.2. Thus, both the intensity and magnitude of human disturbances vary locally throughout the park. However, several committed individuals serving within the Forest Department have worked exceptionally well to establish an effective protection mechanism since the declaration of the National Park (Karanth et  al. 1999). This protection mechanism involved extensive foot patrols during the day, vehicular night patrols, patrolling in boats along the reservoirs, strategic placement of a network of anti-­ poaching camps, a wireless communication system and a host of incentives and welfare schemes for the protection staff on ground (Karanth et al. 2001). This mechanism worked effectively in the first two decades since the declaration of the National Park, resulting in dramatic recovery of the habitat through control of poaching, habitat manipulation, encroachments and habitat resource extraction. Subsequently, the management effectiveness varied locally over the years largely due to frequent changes in leadership and related park governance issues (N. S. Kumar and K. U. Karanth, pers. obs). Thus, although the overall management of the park largely remained effective, particularly when compared with other parks in the state and India, its spatial coverage and efforts varied locally throughout the park.

1.6.2 Bandipur National Park Bandipur was an exclusive hunting reserve for the erstwhile kings of Mysore, and its history dates back to 1931, when a 90  km2 sanctuary was established in the Chamarajanagar State Forest under the Mysore Game and Forest Preservation Regulation. In 1941, the area was expanded to 800 km2 and designated as Venugopala Wildlife Park, named after the deity atop the hill Gopalaswamy Betta, the highest peak (1455 m) in the park. Currently, the park is spread over an area of 880 km2, and it is one of the first nine tiger reserves created under Project Tiger in 1973. Bandipur is located at the heart of an extensive forest at the confluence of the Western Ghats and Nilgiri Hills in the Mysore and Chamrajanagar Districts of Karnataka state. Bandipur is a part of the ecological continuum that includes Nagarahole on the northwest (across the Kabini reservoir), Wayanad Wildlife Sanctuary (in Kerala state) to the southwest and Mudumalai Wildlife Sanctuary (in Tamil Nadu state) to the south. The altitude ranges between 400 and 1455 m above mean sea level. The Bandipur terrain is more undulating than Nagarahole with several rocky hills and flat-topped hills scattered through the northwestern and southwestern regions. The valleys are drained by several rivers, the Kabini, Nugu, Moyar and Mavinahalla, several perennial streams and a large number of seasonal streams.

16

1  Introduction: The Conservation Issue

The Moyar gorge runs nearly for 20 km along the southern boundary of the park and is as deep as 260 m at several places, with the Moyar River flowing through the gorge. The park is flanked by two major reservoirs, Kabini Reservoir to the northwest and Nugu Reservoir to the north. Water resources are further supplemented through a number of natural and man-made pools and tanks. The annual rainfall of 625–1250 mm is mainly bimodal; with heavier rains from the southwest monsoon during late June to September and lighter rains from the northeast monsoon during the October to November period. The rainfall gradient is east-west with the western parts receiving higher rainfall than the eastern counterpart. The underlying rocks are mainly metamorphic (gneiss, quartzite, mica and hornblende schists) with some igneous material such as granite and charnockite appearing as outcrops. The park has two types of soils; clayey black soils of gneissic origin that do not usually favor tree growth, and red loamy soils originating from granites and gneisses that support luxuriant growth of trees. Patches of shale mixed with powdered quartz, which form natural salt-licks for ungulates, are also found in the park. The annual temperature varies between 10° and 35° C. The forests are mostly of the mixed dry deciduous forest type of the Terminalia— Anogeissus—Tectona series. The other main species of this type are Pterocarpus marsupium, Grewia arborea, Bombax malabaricum, Careya arborea. In the northwestern part where the rainfall is higher, moist deciduous forests of Tectona— Dillenea—Lagerstroemia series occur. The other main species occurring here are Rosewood Dalbergia latifolia, Hunalu Terminalia paniculata, Noga Toona ciliata, Albizia spp., Bambusa arundinacea and Dendrocalamus strictus. The undergrowth is dominated by lantana in large parts of the park. Forest vegetation is affected by annually recurring man-made fires varying in intensity and spread. The present vegetation particularly along the northern and eastern boundaries of the park is likely to be a representation of degraded stages of succession with savanna-woodland to stunted discontinuous scrub thickets (Pascal 1982). Bandipur supports an impressive array of wildlife; 28 species of large mammals are known to occur (Karanth et al. 2001) that include elephant Elephas maximus, gaur Bos frontalis, sambar Cervus unicolor, chital Axis axis, muntjac Muntiacus muntjak, chousingha Tetracerus quadricornis, wild pig Sus scrofa, hanuman langur Semnopithecus entellus and bonnet macaque Macaca radiata. In addition to this, blackbuck Antelope cervicapra and striped hyena Hyaena hyaena occur occasionally on its eastern fringes. The tiger Panthera tigris, leopard Panthera pardus, Asiatic wild dog or dhole Cuon alpinus and sloth bear Melursus ursinus are the large carnivores. Bandipur is also rich in avifauna, with more than 200 species of birds and is declared as an “Important Bird Area” by BirdLife International (BirdLife International 2020b). The herpetofauna includes a variety of snakes, lizards, turtles and frogs. Unlike Nagarahole, Bandipur has no human settlements inside the park. All the anthropogenic pressures on the park originate from nearly 200 villages dotting all along the northern and eastern boundaries of the park (Lal et al. 1994). Madhusudan (2005) recorded a livestock density of 236 animals per km2 and documented that the use of forest resources by local communities was driven mainly by market-linked

1.7  Study Species

17

commercial demands rather than by their subsistence needs. As a result, the forest biomass extraction (e.g., grazing, fuelwood extraction, burning along the edges of the park) increased significantly over the years impacting different aspects (e.g., vegetation ecology, soil, forest hydrology) of the ecosystem functioning in Bandipur. This has led to the degradation of dry deciduous forest patches to scrub forests at the fringe areas heavily affected by chronic anthropogenic activities (Mehta et  al. 2008a). The spatial distribution of the villages surrounding the park, human and livestock population size at each village and the distance between village and park boundary are some of the factors that induce local variation in the human disturbance levels across the park. Like Nagarahole, Bandipur also has a long tradition of effective protection and management, particularly against poaching. The establishment of a network of anti-­ poaching camps and regular patrolling schedules coupled with necessary infrastructure in terms of wireless communication systems, patrolling vehicles, boats and other protection infrastructure (Kantharaju 2000) have helped maintain relatively high levels of management effectiveness. The biggest challenge to management in this park has been the control of fire and forest biomass extractive practices. The linear configuration of the park, vast expanses of neighboring forests on its southern side, intense anthropogenic activities on the northern boundaries and the presence of religious shrines inside the park pose different sets of challenges to management in Bandipur. Consequently, there is local variation in the effectiveness of management measures.

1.7

Study Species

Five large ungulate species, differing in body size and diet, were chosen to evaluate their habitat relationships and drivers of their local abundance. The study species were: red muntjac, wild pig, chital, sambar and gaur. The body size of the study species ranged from a 20-kg muntjac to an 1000-kg gaur (Karanth and Sunquist 1992), and they also occupied a wide range of habitat niches ranging from dense forests to open woodland and forest-edge habitats. These ungulates play a key role in forest dynamics by influencing seed dispersal, succession patterns and forest structure (Karanth and Sunquist 1992; Prasad et al. 2006). Macdonald (2001) and, Wilson and Mittermeier (2011) provide detailed accounts of the study species. A brief ecological description of each of the study species is given below.

1.7.1 Wild Pig (Sus scrofa) Wild pig (family Suidae) is a generalist ungulate, and it perhaps has the widest geographical distribution among all terrestrial mammals, occurring in all continents except Antarctica (Keuling and Leus 2019), with as many as 25 recognized subspecies although the number of truly differentiated subspecies is still highly debated (Meijaard et al. 2011). It is widely distributed throughout India (Karanth et al. 2009)

18

1  Introduction: The Conservation Issue

and found in a variety of habitats ranging from semi-deserts to tropical rain forests, woodlands to open grasslands and open habitats, frequently venturing into adjacent agricultural lands for forage. It is a social non-ruminant that lives in small groups and is occasionally found in large and gregarious groups of varying size depending on locality and season. Wild pig body size ranges between 25 and 100 kg. Their food includes roots, tubers, bulbs, bark, flowers, fruits, seeds and sedges that constitute nearly 90% of their diet, as well as carrion and soil invertebrates (Mason 1893; Spitz 1986). They are often considered as effective seed dispersers. Although wild pigs prefer swampy areas, they readily take shelter in thick shrubs and grass thickets in dry forests. Generally, they are most active during early mornings and late afternoons/early evening, but are observed to be nocturnal in their habit in highly disturbed habitats. They need to regularly drink water and may cover considerable distances to do so (Prater 1985). They raid crops regularly and are considered as pests in many countries. Among all the ungulates, wild pig is known to inflict heavy losses to agricultural crops in cultivated areas alongside habitats where it occurs (Prater 1985; Karanth et al. 2013). It is one of the most widely and illegally hunted ungulates in India (Madhusudan and Karanth 2002), and hence wild pig densities are considerably depressed in heavily hunted areas, in spite of its relatively high abundance in many parts across its range. Although prolific breeders, wild pigs are generally found in low densities ranging between 0.4 and 3.6 animals per km2 (Karanth and Nichols 2000; Karanth and Kumar 2005) while their recorded density was 8 animals per km2 in Kawal Wildlife Sanctuary in central India (Siddiqui 2010). Recent surveys in India (Jhala et al. 2015) across potential tiger habitat have recorded similar or higher densities of wild pigs (≥8 animals per km2) in Pench and Kanha (Madhya Pradesh), Corbett (Uttarakhand) and Sariska (Rajasthan) reserves. Wild pig populations are known to undergo frequent fluctuations due to disease outbreaks (Santiapillai and Chambers 1980). Prater (1985) gives a detailed natural history account of the species, while the global conservation status including its distributional range, habits, habitat and threats is provided by Keuling and Leus (2019).

1.7.2 Red Muntjac (Muntiacus muntjak) Muntjac (family Cervidae) is a small (90 kg. Chital thrive in a wide variety of habitats, but prefer open forests and grasslands (Mishra 1982). They are typical inhabitants of the grassland-­forest ecotone (Mattioli 2011) and avoid steep terrain and evergreen forests (Mishra and Wemmer 1987; Karanth et al. 2009). Karanth and Sunquist (1992) found well-­protected mixed forests with teak plantations, swampy grasslands and moist deciduous patches to support high abundance of chital. Chital are preferential grazers (Dinerstein 1979; Mishra and Wemmer 1987). Schaller (1967) found grasses

20

1  Introduction: The Conservation Issue

constitute >60% of their diet in Kanha, Central India. However, chital also readily browse on leaves from a number of trees, shrubs and vines and seasonal fruits (Dinerstein 1979; Johnsingh and Sankar 1991) and are known to feed also on crabs in the Sundarbans (Mattioli 2011). Hofmann (1989) classified chital as an intermediate opportunistic feeder. Chital drink water regularly and are known to be intolerant to heat, preferring to rest in shade. Chital are also an important prey for large predators (Karanth and Sunquist 1995; Andheria et al. 2007) and their fawn mortality rate can be as high as 66% in predator-rich areas (Pariwakam 2006). The ecological densities of chital range from 1 to 51.3 per km2 (Karanth and Nichols 2000; Karanth and Kumar 2005) although their local densities can be as high as 100–120 per km2 (Gangadharan 2005; Kumar 2011). In recent country-wide surveys of tigerprey populations (Jhala et al. 2015), high chital densities (64 per km2) were also recorded at parts of Corbett (Uttarakhand) and Pench (Madhya Pradesh) reserves. Chital is one of the most widely hunted species, and its densities are highly depressed in heavily hunted (Madhusudan and Karanth 2002) and intensively livestock-grazed areas (Madhusudan 2004). They are also highly susceptible to livestock-borne diseases (Schaller 1967), habitat conversions/manipulations and attacks by free-ranging domestic dogs (Raman 2015). However, chital are prolific breeders; empirical observations (Karanth et  al. 1999 in Nagarahole) and population trend data (K. U. Karanth and N. S. Kumar, Unpublished data from Bhadra) show that chital populations can dramatically recover in response to removal of anthropogenic pressures and improved protection measures. Although chital populations are declining drastically outside protected areas, because of their wide distribution, a large number of subpopulations throughout their range and insufficient evidence to meet the ‘near threatened’ status criterion, chital has been assigned the ‘least concern’ conservation status by IUCN (Duckworth et al. 2015). Excellent and detailed accounts of the species are available in Schaller (1967), Raman (2015) and Duckworth et al. (2015).

1.7.4 Sambar (Rusa unicolor) Sambar (family Cervidae) is predominantly a forest ungulate adapted to a variety of habitats throughout India, Nepal, Bhutan, Sri Lanka, Bangladesh, South China, mainland southeast Asia and many of the main islands of Indonesia (Khan and Johnsingh 2015). However, nearly one-third of its distributional range in India has contracted in the past 100 years (Karanth et al. 2010), and its current distribution is highly fragmented and under rapid decline in the rest of its range. In India, large populations of sambar exist only in secured and well-protected reserves (Karanth et  al. 2009). Its body size varies between 100 and 250  kg and fully-grown adult stags may weigh ~320 kg. The ecological densities of sambar range from 1.5 to 10.7 per km2 (Karanth and Nichols 2000; Karanth and Kumar 2005) depending on protection efficacy. Recent surveys of prey populations in potential tiger habitats in India (Jhala et al. 2015) have recorded high densities of sambars in well protected parts of Rajaji (12 per km2), Sariska (14 per km2), Pakke (16 per km2) and

1.7  Study Species

21

Ranthambore (26 per km2) reserves. They occur in very low densities in poorly protected and highly disturbed forested habitats and in alluvial grasslands of Kaziranga. Although sambar is a non-social ruminant, family associations consisting of 3–5 individuals are common. Their group sizes appear to be larger in dry deciduous and semi-arid regions. Temporary association in large numbers, often consisting of 60–100 individuals, is common near well-wooded large waterbodies during summer. Sambars are mostly crepuscular and their activity pattern seems to shift to nocturnal habits in disturbed areas. They are selectively preyed upon by large carnivores (Karanth and Sunquist 1995) and are a dominant source of diet for sympatric carnivores (Karanth and Kumar 2005; Andheria et al. 2007). Sambar is a large intermediate selective feeder (Hofmann 1989) that prefers dense cover and steep terrain. Their day time resting sites often include deep valleys in highly undulating hills. Schaller (1967) found Sambar to subsist on a wide variety of plants in central India. Sambar is also known to feed on various browse such as twigs, bark, wild fruits, shoots and leaves of trees. Its broad diet is perhaps a consequence of its habitat flexibility. It is known to drink water regularly, and it typically lies up in thickets, when it is not foraging. Because of their preference for undulating terrains and water, sambars are usually abundant in well-watered, well-protected undulating forest patches. Hunting or poaching and habitat encroachment/loss are the main threats to sambar. Sambar is also an occasional crop pest throughout its range. Although, effects of livestock grazing are minimal on sambar because of its browsing habits, lopping and fuelwood removal appear to deprive browsing resources for sambars. In many places across Southeast Asia, they have been hunted to local extinction, and their meat is commonly available in local markets. There are several reports of poaching instances of sambar even in high-profile and reasonably well-­ protected reserves of India. They are very sensitive to disturbance and forest resource extractive activities. However, sambar populations can recover quickly if the anthropogenic pressures are removed and protection measures improved (Karanth et  al. 1999; Karanth, K. U. and N. S. Kumar, Unpublished data). The ongoing collapse of sambar populations in many parts of its range in Southeast Asia indicates that sambar is a conservation dependent species, and a network of protected areas is critical for sustaining its populations. Sambar is declared as a ‘vulnerable’ species by IUCN because of its sustained decline throughout its range, particularly during the last three decades. Excellent accounts of the species and its conservation status are available in Khan and Johnsingh (2015) and Timmins et al. (2015).

1.7.5 Gaur (Bos frontalis) Gaur (family Bovidae) is arguably the largest bovid species in Asia with adult body masses between 500 and 900 kg, although large bulls may exceed 1000 kg. Although its historic distribution ranged throughout mainland South and Southeast Asia and Sri Lanka, the current distribution of gaur is highly scattered and fragmented in India, Lao PDR, Myanmar, China and Malaysia, while it is extinct in Sri Lanka and Bangladesh. Gaur was once found widely in the forests of central and south India,

22

1  Introduction: The Conservation Issue

but now has one of the most restricted distributional ranges among herbivores (Karanth et  al. 2009), with an estimated range contraction of ~60% in the past 50 years (Karanth et al. 2010). Its local extinction rate in India, a stronghold of gaur distribution, has particularly ranged between 7% in reasonably protected habitats to 98% in unprotected habitats (Karanth et al. 2010), with a reduction of 80% in the past 100 years in its global distribution. Its population in India is now reduced to four large clusters (Western Ghats, Eastern Ghats, Central Indian and Northeast forests) and two minor clusters (Bihar and West Bengal) of populations. The most extensive stronghold of gaur is in the Western Ghats, and nearly 60% of the 22,000 km2 landscape in the state of Karnataka is occupied by gaur (K. U. Karanth and N. S. Kumar, Unpublished data). The ecological densities of gaur vary from 0.2 to 11.3 per km2 (Karanth and Nichols 2000; Karanth et al. 2001, 2008; Karanth and Kumar 2005; Rayar 2010; Jhala et al. 2015). Gaur is not found in arid and semi-arid forests (Karanth et al. 2009), while it is in extremely low numbers in alluvial floodplains of Brahmaputra and rainforests of Arunachal Pradesh in India (Karanth and Nichols 2000). Gaur populations are known to fluctuate periodically even in well-­ protected reserves such as Nagarahole, Bandipur and Bhadra, due to disease outbreaks, seasonal movements and other reasons (K. U. Karanth and N. S. Kumar, Unpublished data). Madhusudan and Karanth (2002) found gaur densities severely depressed in more hunted areas than in less hunted areas of Nagarahole, although both areas shared similar habitat characteristics. In several sites across Southeast Asia, local populations of gaur became extinct due to heavy hunting pressures. Gaur is predominantly a forest grazer and a bulk feeder (Hofmann 1989). Although several studies indicate gaur preference for hilly terrains, it is not uncommon in lowland plains. Because most of the forest plains are either heavily disturbed or converted to croplands by humans, it is possible that such large scale habitat modifications and disturbances might have resulted in the apparent preference for hilly habitats (Schaller 1967). It lives in social groups, although adult bull gaurs are usually solitary. The basic social group consists of several females and juveniles. It is common to see temporary aggregations of gaur numbering about 20–50  in large feeding groups during monsoon season. It is primarily a grass-roughage eater and its diet includes: coarse and dry grasses, bamboo, leaves and twigs of shrubs, forbs and trees (Sankar et al. 2015). It occasionally feeds on tree bark as well (Pasha et al. 2002; Ahrestani 2009). Gaurs are mostly diurnal and crepuscular in undisturbed habitats, typically lying up in forest clearings. Their habits seem to turn nocturnal in heavily disturbed sites. Gaurs move widely compared to other ungulates, and their relatively large scale seasonal movements in response to food resource availability have been frequently observed. They are known to drink water at least once a day (Schaller 1967). Gaurs are selectively preyed by tigers in areas where prey of different body-size classes are abundantly available (Karanth and Sunquist 1995). Gaur together with Sambar contributed to nearly 70–80% of the prey biomass consumed by tigers in central Indian and Western Ghats forests (Karanth and Sunquist 1995; Karanth and Kumar 2005; Andheria et al. 2007). Predation and disease appear to be main causes of gaur mortality. Gaur populations can undergo severe fluctuations due to disease outbreaks. Hunting (for meat in South Asia and for commercial trade in Southeast Asia), habitat loss, habitat conversion/modification and habitat

1.9  Organization of the Monograph

23

degradation are the main threats to gaur populations. They are also vulnerable to disease outbreaks and very sensitive to anthropogenic activities. Gaur densities were the most depressed among all the ungulates in areas heavily grazed by livestock (Madhusudan 2004) and where forest biomass extraction was linked to commercial market demands (Madhusudan 2005). However, their populations in many parts of India have recovered well, when anthropogenic pressures were removed through positive conservation interventions (Karanth et al. 1999; K. U. Karanth and N. S. Kumar, Unpublished data). Gaur is also a conservation-dependent species, as they seem to be benefitted by the network of protected areas where protection effectiveness is both de facto and de jure. The overall conservation status of the species has been assessed as ‘vulnerable’ by the IUCN due to a drastic decline of more than 30% over past three generations, and its current distribution is mostly limited to habitats within protected areas (Duckworth et al. 2016). India appears to be gaur’s last stronghold with its presence in 15 states (Groves and Leslie Jr 2011). Excellent species accounts can be found in Ahrestani and Karanth (2014), and its detailed conservation assessment in Duckworth et al. (2016).

1.8

Ecological Comparisons

The five study species represent four body-size classes: Small ungulates of 20–30 kg body mass, medium sized ungulates with body size varying between 30 and 100 kg, large ungulates of about 100–350 kg of body mass and ‘mega ungulates’ of >500 kg of body mass. They occupy a wide variety of habitats from forest-edge to open broken canopy forests to dense woodland habitats (Macdonald 2001; Wilson and Mittermeier 2011). Even though all the study species prefer green grasses when available, their dietary preferences vary in periods of food scarcity in hot and dry seasons (Hofmann 1989). While muntjac is highly selective in its food habits (Ilyas and Khan 2003), chital prefers grass to a greater extent (Dinerstein 1979) and sambar and gaur seem to be least specialized mixed feeders (Hofmann 1989; Wilson and Mittermeier 2011). Wild pigs (Spitz 1986) are extreme generalists that can readily adapt to almost any available food. Terrain, water, shade and absence of human disturbance seem to play an important role in the habitat use of all the study ungulates. Thus, the five ungulate species we have chosen to study in this monograph provide good examples to investigate the relative importance of ecological factors in shaping their distribution and population densities at both local and landscape levels. Furthermore, all these species are widely distributed across India and in some cases the rest of tropical Asia, and the findings of this study would be relevant to conservation and species recovery at many other sites.

1.9

Organization of the Monograph

This monograph addresses spatial variation in the patterns of ungulate abundance and distribution to understand key factors that determine their abundances. In this chapter (Chap. 1: Introduction), we have provided the conservation context in which

24

1  Introduction: The Conservation Issue

the examination of ungulate-habitat relationships has become necessary for informed management. We have also explained both ecological and methodological challenges involved in undertaking such an investigation. Background information on the study landscape and a synthesis of current knowledge on biology and conservation issues affecting the study species will enable formulation of a priori hypotheses on the potential drivers of ungulate abundance in tropical deciduous forest systems. This chapter has, thus, identified the key objectives and the central hypotheses to be tested in the study. The second part of the monograph (Chap. 2) focuses on methodological development: the formulation of appropriate statistical models representing the hypothesized ungulate-habitat relationships, under the overarching methodological framework of a hierarchical modeling approach. In Chap. 2, relying on this modeling philosophy, we formulate explicit spatial models to represent the likely mechanisms governing both the biological and sampling processes that generate the data we analyze. We demonstrate how a simple, flexible, yet robust modeling structure can be used to reliably assess species-habitat relationships without ignoring sampling biases. This chapter also explains a Bayesian model selection approach developed by Kuo and Mallick (1998), which is used to assess the relative importance of predictor variables in influencing observed abundance patterns. We include in this chapter implementation of all these statistical models in readily accessible free software packages such as NIMBLE (de Valpine et al. 2017) in R (R Core Team 2019). We illustrate application of these models by confronting model-based predictions about local abundance of chital deer against data obtained from line transect surveys in the NagaraholeBandipur region. In the third and central part of the monograph (Chap. 3), we examine ungulate-­ habitat relationships for all five study species using the models developed in Chap. 2. First, we identify and explain the rationale behind the choice of a set of predictor variables characterizing the ecological and management system that could potentially set ungulate abundance levels. Thereafter, we formulate species-specific a priori hypotheses and then confront them with data from line transect surveys on these ungulates. Methods including survey design and field protocols that were followed for collecting data on ungulates and environmental covariates are also included in this chapter. Using models formulated in Chap. 2, we examine both the individual and combined effects of different predictor variables on local population densities of each species. We then compare the relative importance of different ecological factors associated with observed abundance patterns. In the final part of the monograph (Chap. 4), we evaluate the ecological and management implications of the ungulate-habitat associations derived in Chap. 3. We specifically examine the influence of human impacts on even relatively better protected landscapes such as the Nagarahole-Bandipur study landscape. We explain how human activities have a pervasive influence, whether they reflect local consumption or commercial exploitation or both. In Chap. 4, we also discuss the imperative for a systematic assessment of the human use of the ungulate habitat within a reliable sampling framework. The patterns of these human impacts are then examined by mapping their spatial distribution and relative intensity to identify specific

References

25

vulnerability zones for ungulate populations in the study landscape. We also discuss how a systematic line transect survey design can be effectively used to monitor proximate threats to ungulate populations and their habitats, across space and time. We then identify a set of practical management actions to enhance ungulate abundance levels and demonstrate the need for and utility of setting up objective biological targets for evaluation of management effectiveness to achieve population recoveries of threatened ungulate populations across Asia’s tropical forests. In the concluding chapter (Chap. 5), we present an overall perspective of ungulate conservation and discuss the way forward to achieve conservation goals. Here, we outline the limitations, as well as utility, of the results for management and conservation of ungulates in tropical deciduous forests of India and potentially, the rest of tropical Asia. We specifically discuss how the same sets of ecological and management related variables induce different effects on different ungulates depending on their ecological traits, thereby emphasizing the need for informed management interventions. We also describe the basic tenets of adaptive management and the steps involved in structured decision making processes for managing ungulate populations. We illustrate the ability of the overarching hierarchical modeling and estimation approach used in this study to facilitate adaptive management of the ungulate populations and their habitats in challenging environments.

References Adler PB, Raff DA, Laurenroth WK (2001) The effect of grazing on the spatial heterogeneity of vegetation. Oecologia 128:465–479 Ahrestani FS (2009) Asian eden—large herbivore ecology in India. PhD Thesis, Wageningen University, The Netherlands Ahrestani FS, Karanth KU (2014) Gaur (Bos gaurus C. H. Smith, 1827). Species accounts. In: Melletti M, Burton J (eds) Ecology, evolution and behaviour of wild cattle: implications for conservation. Cambridge University Press, Cambridge, pp 174–193 Ahrestani FS, Sankaran M (2016) Introduction: the large herbivores of South and Southeast Asia—a prominent but neglected guild. In: Ahrestani FS, Sankaran M (eds) The ecology of large herbivores in South and Southeast Asia. Ecological Studies, vol 225. Springer Science, pp 1–13. https://doi.org/10.1007/978-94-017-7570-0_1 Albert A, Marell A, Picard M, Baltzinger C (2015) Using basic plant traits to predict ungulate seed dispersal potential. Ecography 38:440–449. https://doi.org/10.1111/ecog.00709 Anderson VJ, Briske DD (1995) Herbivore-induced species replacement in grasslands: is it driven by herbivory tolerance or avoidance? Ecol Appl 5:1014–1024 Andheria A, Karanth KU, Kumar NS (2007) Diet and prey profiles of three sympatric large carnivores in Bandipur Tiger Reserve in India. J Zool 273:169–175 Appayya MK (2001) Management plan for Rajiv Gandhi (Nagarahole) National Park (2000–2010). Karnataka Forest Department, Government of Karnataka, Mysore Archer S (1995) Herbivore mediation of grass-woody plant interactions. Trop Grassl 29:218–235 Augustine DJ, McNaughton SJ (2006) Interactive effects of ungulate herbivores, soil fertility, and variable rainfall on ecosystem processes in a semi-arid savanna. Ecosystems 9:1242–1256 Bagchi S, Goyal SP, Sankar K (2004) Herbivore density and biomass in a semi-arid tropical dry deciduous forest of western India. J Trop Ecol 20:475–478 Baillie JEM, Hilton-Taylor C, Stuart S (eds) (2004) 2004 IUCN red list of threatened species: a global assessment. IUCN, Gland

26

1  Introduction: The Conservation Issue

Bardgett RD, Wardle DA (2003) Herbivore-mediated linkages between above-ground and below-­ ground communities. Ecology 84:2258–2268 Barrette C (1977) Some aspects of the behavior of muntjacs in Wilpattu National Park. Mammalia 41:1–34 Barrette C (2015) Muntjac or barking deer Muntiacus muntjac. In: Johnsingh AJT, Manjrekar N (eds) Mammals of South Asia, vol 2. Universities Press (India) Private Limited, New Delhi, pp 176–191 Barve N, Kiran MC, Vanaraj G, Aravind NA, Rao D, Shaanker RU, Ganeshaiah KN, Poulsen JG (2005) Measuring and mapping threats to a wildlife sanctuary in Southern India. Conserv Biol 19:122–130 Bawa KS, Das A, Krishnaswamy J, Karanth KU, Kumar NS, Rao M (2007) Ecosystem profile: Western Ghats and Sri Lanka biodiversity hotspot—Western Ghats region. Critical Ecosystems Partnership Fund, Arlington, VA Berwick SH (1974) The community of wild ruminants in the Gir forest ecosystem, India. PhD dissertation, Yale University, USA Bibi F, Metais G (2016) Evolutionary history of the large herbivores of South and Southeast Asia (Indomalayan realm). In: Ahrestani FS, Sankaran M (eds) The ecology of large herbivores in South and Southeast Asia. Ecological Studies, vol 225, Springer Science. pp 15–88. https://doi. org/10.1007/978-94-017-7570-0_2 BirdLife International (2020a) Important bird areas factsheet: Nagarhole National Park. http:// www.birdlife.org. Accessed 31 Jan 2020 BirdLife International (2020b) Important bird areas factsheet: Bandipur National Park. http:// www.birdlife.org. Accessed 31 Jan 2020 Boulanger V, Baltzinger C, Saïd S, Ballon P, Ningre F, Picard J, Dupouey J (2011) Deer-mediated expansion of a rare plant species. Plant Ecol 212:307–314 Bourgeon G (1989) Explanatory booklet on the reconnaissance soil map of forest area: Karnataka and Goa. French Institute, Pondicherry Briggs J (2003) The biogeographic and tectonic history of India. J Biogeogr 30:381–388 Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers DL, Thomas L (2001) Introduction to distance sampling: estimating abundance of biological populations. Oxford University Press, Oxford Chase JM, Leibold MA, Downing AL, Shurin JB (2000) The effects of productivity, herbivory and plant species turnover in grassland food webs. Ecology 81:2485–2497 Chellam R (1993) The ecology of Asiatic lion Panthera leo persica. PhD thesis, Saurashtra Unversity, India Coe MJ (1983) Large herbivores and food quality. In: Lee JA, McNeill J, Rorison IH (eds) Nitrogen as an ecological factor: the 22nd symposium of the British Ecological Society, Oxford, 1981. Blackwell, Oxford, pp 345–368 Corlett RT (2007) The impact of hunting on the mammalian fauna of tropical Asian forests. Biotropica 39:292–303 Das A, Krishnaswamy J, Bawa KS, Kiran MC, Srinivas V, Kumar NS, Karanth KU (2006) Prioritization of conservation areas in the Western Ghats, India. Biol Conserv 133:16–31 de Valpine P, Turek D, Paciorek CJ, Anderson-Bergman C, Temple Lang D, Bodik R (2017) Programming with models: writing statistical algorithms for general model structures with NIMBLE. J Comput Graph Stat 26:403–413. https://doi.org/10.1080/10618600.2016.1172487 Dinerstein E (1979) An ecological survey of the Royal Karnali-Bardia Wildlife Reserve, Nepal. Part II; Habitat/animal interaction. Biol Conserv 16:265–300 Dinerstein E (1980) An ecological survey of the Royal Karnali-Bardia wildlife reserve, Nepal. Part III: ungulate populations. Biol Conserv 18:5–37 Dinerstein E (2003) The return of the unicorns: the natural history and conservation of the greater one-horned rhinoceros. Columbia University Press, New York Dinerstein E, Loucks C, Wikramanayake E, Ginsberg J, Sanderson E, Seidensticker J, Forrest J, Bryja G, Heydlauff A, Klenzendorf S, Leimgruber P, Mills J, O’Brien TG, Shrestha M, Simons R, Songer M (2006) The fate of wild tigers. Bioscience 57:508–514

References

27

Directorate of Census Operation (2004) Primary census abstract for Andhra Pradesh, Karnataka & Lakshadweep. Technical Report Vol. X. 00-063-2001-Cen-CD.Office of the Registrar General and Census Commissioner, Government of India, New Delhi Dormann CF, McPherson JM, Araujo MB, Bivand R, Bolliger J, Carl G, Davies RG, Hirzel A, Jetz W, Kissling WD, Kuhn I, Ohlemuller R, Peres-Neto PR, Reineking B, Schroder B, Schurr FM, Wilson R (2007) Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30:609–628 Du Toit JT, Owen-Smith N (1989) Body size, population metabolism and habitat specialization among large African herbivores. Am Nat 133:736–740 Duckworth JW, Kumar NS, Islam MA, Baral HS, Timmins R (2015) Axis axis. The IUCN red list of threatened species 2015, p e.T41783A22158006. https://doi.org/10.2305/IUCN.UK.2015-4. RLTS.T41783A22158006.en Duckworth JW, Sankar K, Williams AC, Kumar NS, Timmins RJ (2016) Bos gaurus. The IUCN red list of threatened species 2016, p e.T2891A46363646. https://doi.org/10.2305/IUCN. UK.2016-2.RLTS.T2891A46363646.en Duncan P, Danell K, Bergstrom R, Pastor J (2006) Introduction. In: Danell K, Duncan P, Bergstrom R, Pastor J (eds) Large herbivore ecology, ecosystem dynamics and conservation. Cambridge University Press, Cambridge, pp 1–10 East R (1984) Rainfall, soil nutrient status and biomass of large African savanna mammals. Afr J Ecol 22:245–270 Eisenberg JF (1980) The density and biomass of tropical mammals. In: Soule ME, Wilcox BA (eds) Conservation biology: an evolutionary-ecological perspective. Sinauer Associates, Sunderland, pp 34–55 Eisenberg JF, Lockhart H (1972) An ecological reconnaissance of Wilpattu National Park, Ceylon. Smithson Contrib Zool 101:1–118 Eisenberg JF, Seidensticker J (1976) Ungulates in southern Asia: a consideration of biomass estimates for selected habitats. Biol Conserv 10:293–308 Frank DA, McNaughton SJ, Tracy BF (1998) The ecology of the earth’s grazing ecosystems: comparing the Serengeti and Yellowstone. Bioscience 48:513–521 Frank DA, Kuns M, Guido D (2002) Consumer control of grassland plant production. Ecology 83:602–606 Ganesh T, Devy S, Aravind, Rao D (2005) Assessment of biodiversity of lesser-known and functionally important groups in Nagarhole National Park: ants (Hymenoptera: Formicidae) and dung beetles (Coleoptera: Scarabaeidae). Eco-development Project, Karnataka Forest Department, Hunsur Gangadharan A (2005) Density estimation and time trend analysis of large herbivores in Nagarahole, India. MSc dissertation, University of St. Andrews, UK Groom MJ (2006) Threats to biodiversity. In: Groom MJ, Meffe GK, Carroll CR (eds) Principles of conservation biology. Sinauer Associates, Sunderland, pp 63–109 Groves C (2003) Taxonomy of ungulates of the Indian subcontinent. J Bombay Nat Hist Soc 100(2–3):341–361 Groves CP, Leslie DM Jr (2011) Family bovidae (hollow-horned ruminants). In: Wilson DE, Mittermeier RA (eds) Handbook of the mammals of the world. Hoofed Mammals, vol 2. Lynx Edicions, Barcelona, pp 444–779 Haining RP (1990) Spatial data analysis in the social and environmental sciences. Cambridge University Press, Cambridge Haining RP (2003) Spatial data analysis: theory and practice. Cambridge University Press, Cambridge Harihar A, Pandav B, Goyal SP (2009) Responses of tiger (Panthera tigris) and their prey to removal of anthropogenic influences in Rajaji National Park, India. Eur J Wildl Res 55(2):97–105 Harihar A, Pandav B, MacMilllan DC (2014) Identifying realistic recovery targets and conservation actions for tigers in a human-dominated landscape using spatially explicit densities of wild prey and their determinants. Divers Distrib 20:567–578

28

1  Introduction: The Conservation Issue

Hiremath AJ, Sundaram B (2005) The fire-lantana cycle hypothesis in Indian forests. Conserv Soc 3:26–42 Hofmann RR (1989) Evolutionary steps of ecophysiological adaptation and diversification of ruminants: a comparative view of their digestive system. Oecologia 78:443–457 Hudson RJ (1984) Body size, energetic and adaptive radiation. In: Hudson RJ, White RG (eds) Bioenergetics of wild herbivores. CRC Press, Boca Raton, pp 1–24 Ilyas O, Khan JA (2003) Food habits of barking deer (Muntiacus muntjak) and goral (Naemorheadus goral) in Binzar Wildlife Sanctuary, India. Mammalia 67:521–531 Islam MZ, Rahmani AR (2004) Important bird areas in India: priority sites for conservation. Indian Bird Conservation Network, Bombay Natural History Society (India) and BirdLife International (UK) IUCN (2019) The IUCN red list of threatened species. Version 2019–3. http://www.iucnredlist.org Jathanna D, Karanth KU, Johnsingh AJT (2003) Estimation of large herbivore densities in the tropical forests of southern India using distance sampling. J Zool 261:285–290 Jhala YV, Qureshi Q, Gopal R (eds) (2015) The status of tigers, copredators and prey in India 2014. National Tiger Conservation Authority, New Delhi and Wildlife Institute of India, Dehradun. TR 2015/021 Johnsingh AJT (1983) Large mammalian prey-predators in Bandipur. J Bombay Nat Hist Soc 80:1–57 Johnsingh AJT, Sankar K (1991) Food plants of chital, sambar and cattle on Mundanthurai plateau, Tamil Nadu, South India. Mammalia 55:57–66 Joseph LN, Elkin C, Martin TG, Possingham HP (2009) Modeling abundance using N-mixture models: the importance of considering ecological mechanisms. Ecol Appl 19:631–642 Kantharaju HC (2000) Management plan of Bandipur Tiger Reserve (2000–2005). Karnataka Forest Department, Government of Karnataka, Project Tiger Division, Bandipur Karanth KU, Karanth KK (2007) Free to move: conservation and voluntary resettlements in the Western Ghats of Karnataka, India. In: Redford KH, Eva Fearn (eds) Protected areas and human displacement: a conservation perspective. WCS Working Paper No. 29. Wildlife Conservation Society, New York Karanth KU, Kumar NS (2001) As good as it gets: large herbivore densities in south Asian forests. In: Ganeshaiah KN, Uma Shaanker R, Bawa KS (eds) Tropical ecosystems: structure, diversity and human welfare. Proceedings of the international conference on tropical ecosystems. Oxford-IBH, New Delhi, pp 489–492 Karanth KU, Kumar NS (2005) Distribution and dynamics of tiger and prey populations in Maharashtra, India. Final technical report. Centre for Wildlife Studies, Bangalore, India Karanth KU, Nichols JD (2000) Ecological status and conservation of tigers in India. Final technical report to US fish and wildlife service—Division of International Conservation and Wildlife Conservation Society. Centre for Wildlife Studies, Bangalore Karanth KU, Sunquist ME (1992) Population structure, density and biomass of large herbivores in the tropical forests of Nagarahole, India. J Trop Ecol 8:21–35 Karanth KU, Sunquist ME (1995) Prey selection by tiger, leopard and dhole in tropical forests. J Anim Ecol 64:439–450 Karanth KU, Sunquist ME, Chinnappa KM (1999) Long term monitoring of tigers: lessons from Nagarahole. In: Seidensticker J, Christie S, Jackson P (eds) Riding the tiger: tiger conservation in human dominated landscapes. Cambridge University Press, Cambridge, pp 114–122 Karanth KU, Bhargav P, Kumar NS (2001) Karnataka tiger conservation project. Final report to save the tiger fund—National Fish and Wildlife Foundation, ExxonMobil corporation and other donors. Wildlife Conservation Society, International Programs, Bronx, NY-10460-1099, USA Karanth KU, Nichols JD, Kumar NS, Link WA, Hines JE (2004) Tigers and their prey: predicting carnivore densities from prey abundance. Proc Nat Acad Sci USA 101:4854–4858 Karanth KU, Nichols JD, Kumar NS, Hines JE (2006) Assessing tiger population dynamics using photographic capture-recapture sampling. Ecology 87:2925–2937

References

29

Karanth KU, Kumar NS, Gopalaswamy AM, Srinivas V (2008) Distribution and dynamics of tiger and prey populations in Karnataka. Final technical report. Wildlife Conservation Society— India Program, Centre for Wildlife Studies, Bangalore, India Karanth KK, Nichols JD, Hines JE, Karanth KU, Christensen NL (2009) Patterns and determinants of mammal species occurrence in India. J Appl Ecol 46:1189–1200 Karanth KK, Nichols JD, Karanth KU, Hines JE, Christensen NL (2010) The shrinking ark: patterns of large mammal extinctions in India. Proc Royal Soc Biol Sci 227:1971–1979 Karanth KK, Gopalaswamy AM, Prasad PK, Dasgupta S (2013) Patterns of human-wildlife conflicts and compensation: insights from Western Ghats protected areas. Biol Conserv 166:175–185 Keitt TH, Bjornstad ON, Dixon PM, Citron-Pousty S (2002) Accounting for spatial pattern when modeling organism-environment interactions. Ecography 25:616–625 Kéry M, Royle JA (2016) Applied hierarchical modeling in ecology: analysis of distribution, abundance and species richness in R and BUGS, vol 1. Academic Press, Cambridge Keuling O, Leus K (2019) Sus scrofa. The IUCN red list of threatened species 2019, p e.T41775A44141833. https://doi.org/10.2305/IUCN.UK.2019-3.RLTS.T41775A44141833.en Khan JA, Johnsingh AJT (2015) Sambar Rusa unicolor. In: Johnsingh AJT, Manjrekar N (eds) Mammals of South Asia, vol. 2. Universities Press (India) Private Limited, pp 223–241 Khan JA, Chellam R, Rodgers WA, Johnsingh AJT (1996) Ungulate densities and biomass in the tropical dry deciduous forests of Gir, Gujarat, India. J Trop Ecol 12:149–162 Krishnaswamy J, Bawa KS, Ganeshaiah KN, Kiran MC (2009) Quantifying and mapping biodiversity and ecosystem services: utility of a multi-season NDVI based Mahalanobis distance surrogate. Remote Sens Environ 113:857–867 Kumar NS (2000) Ungulate density and biomass in the tropical semi-arid forest of Ranthambore, India. Master’s thesis, Salim Ali School of Ecology and Environmental Sciences, Pondicherry University, India Kumar NS (2011) Assessment of distribution and abundance of ungulate prey using spatial models in Nagarahole and Bandipur tiger reserves of India. PhD thesis, Manipal University, India Kumar R, Shahabuddin G (2006) Effects of biomass extraction on vegetation structure, diversity and composition of forests in Sariska Tiger Reserve, India. Environ Conserv 32:1–12 Kumar A, Wright B (1999) Combating tiger poaching and illegal wildlife trade in India. In: Seidensticker J, Christie S, Jackson P (eds) Riding the Tiger: Tiger conservation in human dominated landscapes. Cambridge University Press, Cambridge, pp 243–251 Kuo L, Mallick B (1998) Variable selection for regression models. Sankhya Indian J Stat 60:65–81 Lal R, Kothari A, Pande P, Singh S (1994) Directory of National Parks and Sanctuaries in Karnataka: Management Status and Profiles. Indian Institute of Public Administration, New Delhi, India Legendre P (1993) Spatial autocorrelation: trouble or new paradigm? Ecology 74:1659–1673 Levin SA (1992) The problem of pattern and scale in ecology: the Robert H. MacArthur Award Lecture. Ecology 73:1943–1967 Lichstein JW, Simons TR, Shriner SA, Franzreb KE (2002) Spatial autocorrelation and autoregressive models in ecology. Ecol Monogr 72:445–463 Link WA, Barker RJ (2010) Bayesian inference with ecological applications. Academic Press, San Diego CA Macdonald DW (ed) (2001) The new encyclopedia of mammals. Oxford University Press, Oxford Madhusudan MD (2004) Recovery of wild large herbivores following livestock decline in a tropical Indian wildlife reserve. J Appl Ecol 41:858–869 Madhusudan MD (2005) The global village: linkages between international coffee markets and grazing by livestock in a south Indian wildlife reserve. Conserv Biol 19:411–420 Madhusudan MD, Karanth KU (2000) Hunting for an answer: local hunters and large mammal conservation in India. In: Robinson JG, Bennett EL (eds) Hunting for sustainability in tropical forests. Columbia University Press, New York, pp 339–355 Madhusudan MD, Karanth KU (2002) Local hunting and the conservation of large mammals in India. Ambio 31:49–54

30

1  Introduction: The Conservation Issue

Madhusudan MD, Mishra C (2003) Why big, fierce animals are threatened: conserving large mammals in densely populated landscapes. In: Saberwal V, Rangarajan M (eds) Battles over nature: science and the politics of conservation. Permanent Black, New Delhi, pp 31–55 Manikandan R, Lakshminarasimhan P (2012) Flowering plants of Rajiv Gandhi (Nagarahole) National Park, Karnataka, India. Check List 8(6):1052–1084. ISSN 1809-127X. www.checklist.org.br Martin C (1977) Status and ecology of the Barasingha (Cervus duvauceli branderi) in Kanha National Park (India). J Bombay Nat Hist Soc 74:60–132 Mason JM (1893) Food of wild boar. J Bombay Nat Hist Soc 8:447 Mattioli S (2011) Family cervidae (deer). In: Wilson DE, Mittermeier RA (eds) Handbook of the mammals of the world. Hoofed Mammals, vol 2. Lynx Edicions, Barcelona, pp 350–443 McCullagh P, Nelder JA (1989) Generalized linear models. Chapman and Hall, London McNaughton SJ (1985) Ecology of a grazing ecosystem: the Serengeti. Ecol Monogr 55:259–294 McNaughton SJ, Oesterheld M, Frank DA, Williams KJ (1989) Ecosystem-level patterns of primary productivity and herbivory in terrestrial habitats. Nature 341:142–144 McNaughton SJ, Banyikwa FF, McNaughton MM (1997) Promotion of the cycling of diet-­ enhancing nutrients by African grazers. Science 278:1798–1800 Meher-Homji VM (1990) Vegetation types of India in relation to environmental conditions. In: Daniel JC, Serao JS (eds) Conservation in developing countries: problems and prospects. Proceedings of the centenary seminar of BNHS.  Bombay Natural History Society, Oxford University Press, Bombay Mehta VK, Sullivan PJ, Walter MT, Krishnaswamy J, DeGloria SD (2008a) Ecosystem impacts of disturbance in a dry tropical forest in southern India. Ecohydrology 1:149–160 Mehta VK, Sullivan PJ, Walter MT, Krishnaswamy J, DeGloria SD (2008b) Impacts of disturbance on soil properties in a dry tropical forest in southern India. Ecohydrology 1:161–175 Meijaard E, d’Huart JP, Oliver WLR (2011) Family suidae (pigs). In: Wilson DE, Mittermeier RA (eds) Handbook of the mammals of the world. Hoofed Mammals, vol 2. Lynx Edicions, Barcelona, pp 248–291 Mishra HR (1982) The ecology and behavior of chital (Axis axis) in Royal Chitwan National Park, Nepal. PhD dissertation, University of Edinburgh, Edinburgh, UK Mishra HR, Wemmer CM (1987) The comparative breeding ecology of four cervids in Royal Chitwan National Park. In: Wemmer CM (ed) Biology and management of the Cervidae. Smithsonian Institution Press, Washington DC, pp 259–271 Mittermeier RA, Mittermeier CG (2005) Megadiversity: earth’s biologically wealthiest nations. Cemex, Monterrey Nair SSC, Nair PV, Sharatchandra HC, Gadgil M (1977) An ecological reconnaisance of the proposed Jawahar National Park. J Bombay Nat Hist Soc 74:401–435 Neginhal SG (1974) Project tiger. Management plan of the Bandipur Tiger Reserve. Karnataka State. pp 142 Nichols JD, Williams BK (2006) Monitoring for conservation. Trends Ecol Evolut 21:668–673 Oedekoven CS, Buckland ST, Mackenzie ML, King R, Evans KO, Burger LW Jr (2014) Bayesian methods for hierarchical distance sampling models. J Agric Biol Environ Stat 19:219. https:// doi.org/10.1007/s13253-014-0167-0 Olff H, Ritchie ME, Prins HHT (2002) Global environmental controls of diversity in large herbivores. Nature 415:901–904 Owen-Smith N, Mills MGL (2006) Manifold interactive influences on the population dynamics of a multispecies ungulate assemblage. Ecol Monogr 76:73–92 Paine RT (2000) Phycology for the mammalogists: marine rocky shores and mammal-dominated community—how different are the structuring processes. J Mammal 81:637–648 Pariwakam MP (2006) Estimation of abundance and fawn survival in chital (Axis axis) populations using photographic capture-recapture sampling. MSc Thesis, The Manipal Academy of Higher Education, India Pascal JP (1986) Explanatory booklet on the forest map of South India. French Institute, Pondicherry

References

31

Pascal JP, Shyam-Sunder S, Meher-Homji VM (1982) Forest map of South India (Mercara-­ Mysore). French Institute, Pondicherry Pasha MKS, Areendran G, Sankar K, Qureshi Q (2002) Debarking of teak Tectona grandis Linn. F. by gaur Bos gaurus H. Smith during summer in a tropical dry deciduous habitat of central India. J Bombay Nat Hist Soc 99:238–244 Pastor J, Cohen Y (1997) Herbivores, the functional diversity of plant species, and the cycling of nutrients in ecosystems. Theor Popul Biol 51:165–179 Peters RH, Raelson JV (1984) Relations between individual size and mammalian population density. Am Nat 124:498–517 Pollock KH, Nichols JD, Simons TR, Farnsworth GL, Bailey LL, Sauer JR (2002) Large scale wildlife monitoring studies: statistical methods for design and analysis. Environmetrics 13:105–119 Prasad S, Krishnaswamy J, Chellam R, Goyal SP (2006) Ruminant-mediated seed dispersal of an economically valuable tree in Indian dry forests. Biotropica 38:679–682 Prater SH (1985) The book of Indian animals. Bombay Natural History Society, India Pringle RM, Young TP, Rubenstein DI, McCauley DJ (2007) Herbivore-initiated interaction cascades and their modulation by productivity in an African savanna. Proc Nat Acad Sci USA 104:193–197 Puri G, Meher-Homji V, Gupta R, Puri S (1983) Forest ecology—phytogeography and Forest conservation. Oxford IBH, New Delhi Putman R (1988) The natural history of deer. Christopher Helm, London R Core Team (2019) R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/ Raman TRS (2015) Chital Axis axis. In: Johnsingh AJT, Manjrekar N (eds) Mammals of South Asia, Vol. 2. Universities Press (India) Private Limited, pp 192–222 Rayar K (2010) Assessing potential tiger habitats in Cauvery Wildlife Sanctuary, Karnataka using occupancy modelling approaches. MSc dissertation, Tata Institute of Fundamental Research, National Centre for Biological Sciences, India Reddy GV, Karanth KU, Kumar NS, Krishnaswamy J, Karanth KK (2016) Recovering biodiversity in Indian forests. Springer briefs in ecology. Springer. https://doi.org/10.1007/978-981-10-0911-2 Ripple WJ, Newsome TM, Wolf C, Dirzo R, Everatt KT, Galetti M, Hayward MW, Kerley GIH, Levi T, Lindsey PA, Macdonald DW, Malhi Y, Painter LE, Sandom CJ, Terborgh J, Valkenburgh BV (2015) Collapse of the world’s largest herbivores. Sci Adv 1(4):e1400103. https://doi. org/10.1126/sciadv.1400103 Ritchie ME, Tilman D, Knops JMH (1998) Herbivore effects on plant and nitrogen dynamics in oak savanna. Ecology 79:165–177 Robinson JG, Bennett EL (eds) (2000) Hunting for sustainability in tropical forests. Columbia University Press, New York Royle JA, Dorazio RM (2006) Hierarchical models of animal abundance and occurrence. J Agric Biol Environ Stat 11:249–263 Royle JA, Dorazio RM (2008) Hierarchical modeling and inference in ecology: the analysis of data from populations, metapopulations and communities. Academic Press, San Diego Royle JA, Dawson DK, Bates S (2004) Modeling abundance effects in distance sampling. Ecology 85:1591–1597 Royle JA, Kéry M, Gautier R, Schmid H (2007) Hierarchical spatial models of abundance and occurrence from imperfect survey data. Ecol Monogr 77:465–481 Sanderson EW, Forrest J, Loucks C, Ginsberg JR, Dinerstein E, Seidensticker J, Leimgruber P, Songer M, Heydlauff A, O’Brien T, Bryja G, Klenzendorf S, Wikramanayake ED (2006) Setting priorities for the conservation and recovery of wild tigers: 2005–2015. The technical assessment. WCS, WWF, Smithsonian, and NFWF-STF, New York, Washington DC, USA Sankar K, Pasha MKS, Areendran G, Qureshi Q (2015) Gaur Bos gaurus. In: Johnsingh AJT, Manjrekar N (eds) Mammals of South Asia, vol 2. Universities Press (India) Private Limited, pp 287–299

32

1  Introduction: The Conservation Issue

Sankaran M, Ahrestani FS (2016) The ecology of large herbivores of South and Southeast Asia: synthesis and future directions. In: Ahrestani FS, Sankaran M (eds) The ecology of large herbivores in South and Southeast Asia. Ecological Studies, vol 225, Springer Science, pp 237–249. https://doi.org/10.1007/978-94-017-7570-0_11 Santiapillai C, Chambers MR (1980) Aspects of the population dynamics of the wild pig (Sus scrofa, Linnaeus 1758) in the Ruhana National Park, Sri Lanka. Spixiana (Muenchn) 3:239–250 Schaller GB (1967) The deer and the tiger: a study of wildlife in India. University of Chicago Press, Chicago Schaller GB (1972) The Serengeti lion. University of Chicago Press, Chicago Schipper J, Chanson JS, Chiozza F, Cox NA, Hoffmann M, Katariya V, Lamoreux J, Rodrigues ASL, Stuart SN, Temple HJ, Baillie J, Boitani L, Lacher TE Jr, Mittermeier RA, Smith AT, Absolon D, Aguiar JM, Amori G, Bakkour N, Baldi R, Berridge RJ, Bielby J, Black PA, Blanc JJ, Brooks TM, Burton JA, Butynski TM, Catullo G, Chapman R, Cokeliss Z, Collen B, Conroy J, Cooke JG, da Fonseca GAB, Derocher AE, Dublin HT, Duckworth JW, Emmons L, Emslie RH, Festa-Bianchet M, Foster M, Foster S, Garshelis DL, Gates C, Gimenez-Dixon M, Gonzalez S, Gonzalez-Maya JF, Good TC, Hammerson G, Hammond PS, Happold D, Happold M, Hare J, Harris RB, Hawkins CE, Haywood M, Heaney LR, Hedges S, Helgen KM, Hilton-­ Taylor C, Hussain SA, Ishii N, Jefferson TA, Jenkins RKB, Johnston CH, Keith M, Kingdon J, Knox DH, Kovacs KM, Langhammer P, Leus K, Lewison R, Lichtenstein G, Lowry LF, Macavoy Z, Mace GM, Mallon DP, Masi M, McKnight MW, Medellín RA, Medici P, Mills G, Moehlman PD, Molur S, Mora A, Nowell K, Oates JF, Olech W, Oliver WRL, Oprea M, Patterson BD, Perrin WF, Polidoro BA, Pollock C, Powel A, Protas Y, Racey P, Ragle J, Ramani P, Rathbun G, Reeves RR, Reilly SB, Reynolds JE III, Rondinini C, Rosell-Ambal RG, Rulli M, Rylands AB, Savini S, Schank CJ, Sechrest W, Self-Sullivan C, Shoemaker A, Sillero-­ Zubiri C, De Silva N, Smith DE, Srinivasulu C, Stephenson PJ, van Strien N, Talukdar BK, Taylor BL, Timmins R, Tirira DG, Tognelli MF, Tsytsulina K, Veiga LM, Vié JC, Williamson EA, Wyatt SA, Xie Y, Young BE (2008) The status of the world’s land and marine mammals: diversity, threat, and knowledge. Science 322:225–230 Seidensticker J (1976) Ungulate populations in Chitwan valley, Nepal. Biol Conserv 10:183–210 Shiva Prasad CR, Reddy RS, Sehgal J, Velayutham M (1998) Soils of Karnataka for optimizing land use. Publ. 47b executive summary. NBSS (ICAR), State Soil Survey, Department of Agriculture, Govt. of Karnataka Siddiqui I (2010) Predicting tiger (Panthera tigris) densities based on prey abundance in disturbed forests of Kawal. (MSc thesis in wildlife biology and conservation), National Centre for Biological Sciences, Tata Institute of Fundamental Research, Bangalore India Skidmore AK, Ferwerda JG (2008) Resource distribution and dynamics. In: Prins HHT, van Langevelde F (eds) Resource Ecology: spatial and temporal dynamics of foraging. Springer, Dordrecht, pp 57–77 Spitz F (1986) Current state of knowledge of wild boar biology. Pig News Inf 7:171–175 Steinmetz R, Chutipong W, Seuaturien N, Chirngsaard E, Khaengkhetkarn M (2010) Population recovery patterns of Southeast Asian ungulates after poaching. Biol Conserv 143:42–51 Sunquist ME, Sunquist FC, Karanth KU (1999) Ecology, behaviour and resilience of the tiger and its conservation needs. In: Seidensticker J, Christie S, Jackson P (eds) Riding the Tiger: Tiger conservation in human dominated landscapes. Cambridge University Press, Cambridge, pp 5–18 Tamang KM (1982) The status of the tiger (Panthera tigris) and its impact on principal prey populations in the Royal Chitwan National Park, Nepal. PhD dissertation, Michigan State University, USA Terborgh J, Nunez-Iturri G, Pitman NCA, Valverde FHC, Alvarez P, Swamy V, Pringle EG, Paine CET (2008) Tree recruitment in an empty forest. Ecology 89:1757–1768 Thogmartin WE, Sauer JR, Knutson MG (2004) A hierarchical spatial model of avian abundance with application to cerulean warblers. Ecol Appl 14:1766–1779 Thomas L, Buckland ST, Rexstad EA, Laake JL, Strindberg S, Hedley SL, Bishop JRB, Marques TA, Burnham KP (2010) Distance software: design and analysis of distance sampling surveys for estimating population size. J Appl Ecol 47:5–14

References

33

Timmins R, Kawanishi K, Giman B, Lynam A, Chan B, Steinmetz R, Baral HS, Kumar NS (2015) Rusa unicolor. The IUCN red list of threatened species 2015, p e.T41790A85628124. https:// doi.org/10.2305/IUCN.UK.2015-2.RLTS.T41790A22156247.en Timmins RJ, Steinmetz R, Kumar NS, Islam MA, Baral HS (2016) Muntiacus vaginalis. The IUCN red list of threatened species 2016, p e.T136551A22165292. https://doi.org/10.2305/ IUCN.UK.2016-1.RLTS.T136551A22165292.en Vaidyanathan S, Krishnaswamy J, Kumar NS, Dhanwatey H, Dhanwatey P, Karanth KU (2010) Patterns of tropical forest dynamics and human impacts: views from above and below the canopy. Biol Conserv 143:2881–2890 Walker BH, Langridge JL (1997) Predicting savanna vegetation structure on the basis of plant available moisture (PAM) and plant available nutrients (PAN): a case study from Australia. J Biogeogr 24:813–825 Walston J, Robinson JG, Bennett EL, Breitenmoser U, da Fonseca GAB, Goodrich J, Gumal M, Hunter L, Johnson A, Karanth KU, Leader-Williams N, MacKinnon K, Miquelle D, Pattanavibool A, Poole C, Rabinowitz A, Smith JLD, Stokes EJ, Stuart SN, Vongkhamheng C, Wibisono H (2010) Bringing the tiger back from the brink—the six percent solution. PLoS Biol 8:e1000485. https://doi.org/10.1371/journal.pbio.1000485 Wang SW (2010) Estimating population densities and biomass of ungulates in the temperate ecosystem of Bhutan. Oryx 44:376–382 Wegge P, Odden M, Pokharel CP, Storaas T (2009) Predator–prey relationships and responses of ungulates and their predators to the establishment of protected areas: a case study of tigers, leopards and their prey in Bardia National Park, Nepal. Biol Conserv 142:189–202 Williams BK, Nichols JD, Conroy MJ (2002) Analysis and management of animal populations. Academic Press, San Diego Wilson DE, Mittermeier RA (eds) (2011) Handbook of the Mammals of the world. Hoofed Mammals, vol 2. Lynx Edicions, Barcelona Wilson DE, Reeder DM (eds) (2005) Mammal species of the world, 3rd edn. Johns Hopkins University Press, Baltimore MA

2

Development of Hierarchical Spatial Models for Assessing Ungulate Abundance and Habitat Relationships

Abstract

1. Most studies investigating animal abundances and their environmental drivers rely on animal count data generated by interactions between ecological and spatial processes of interest. However, these counts are strongly affected by imperfect detection inherent to the observation methods and processes. Reliable models of animal abundance patterns, which can simultaneously deal with all these important sources of variation, are rarely employed by investigators in practice. We address this key methodological need by developing a hierarchical model for animal-habitat relationships, which can rigorously investigate abundance patterns by explicitly parameterizing ecological, spatial and observational processes. 2. We use a hierarchical formulation with two model components: one describing the ecological processes that determine ungulate abundance and the other addressing the observation process involved in the field survey employing the distance sampling method. We used a Bayesian Poisson regression model to establish the effects of a set of habitat-related factors (predictor variables) on ungulate abundance (response variable). We included a Gaussian Conditional Autoregressive (CAR) prior to account for spatial interaction effects. The hierarchical model permitted specification of covariate effects on abundances at both local and landscape levels. A standard half-normal detection function was used to model the observation process during line transect field surveys. The observation model included 'cluster size' as an individual covariate additionally affecting the detectability of animal groups counted. We specified a zero-truncated Poisson distribution for modeling variations in cluster size. This model was implemented in the programming language R using the package NIMBLE. 3. We demonstrate the application of our hierarchical spatial model to describe the variation in the local abundance of chital deer (Axis axis) in the Nagarahole— Bandipur landscape, based on line transect data from counts of ungulate species

© The Editor(s) (if applicable) and The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2021 N. S. Kumar et al., Spatial Dynamics and Ecology of Large Ungulate Populations in Tropical Forests of India, https://doi.org/10.1007/978-981-15-6934-0_2

35

36

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

obtained by observers walking along 77 line transect samplers (spatial replicates) across a 1400 km2 study area. The count data were accumulated over six temporal replications under rigid field protocols. These counts and associated distance data were used to estimate chital abundance within each cell of a 1-km2 grid superimposed over the landscape. 4. Habitat features such as forest vegetation type, forage availability, distance to water sources, topography, anthropogenic disturbances and effectiveness of law enforcement were considered important predictors of chital abundance at the local scale. The model fitted the survey data well. Expected values of chital densities were highest at local sites with flat terrain, moist deciduous forests and close access to water sources. Results also showed chital abundance was positively influenced by protection effectiveness. Our model reliably predicted and spatially mapped chital abundance across the larger landscape and identified ‘hotspots’ of higher local densities. These predictions potentially can inform management interventions. 5. The flexible structure of our hierarchical spatial model is useful for reliably understanding ungulate-habitat relations, particularly in a context of sparse count data collected under a wide variety of sampling situations where imperfect detections pose a major problem. The model is of potential utility for investigating the spatio-temporal dynamics of animal populations, which can be assessed using distance sampling, for advancing both science and conservation.

2.1

Introduction

Most studies investigating patterns of animal abundance and their associated drivers typically rely on count data (Williams et al. 2002). Variation in count data (numbers of animals or animal clusters) can be attributed to three primary sources: (1) the ecological process where animals respond differently to habitat conditions (Brown et al. 1995), (2) the observation process where detectability of the animals is affected by several factors including animal cluster size, observer’s efficiency, screening effect of vegetation, etc. (Buckland et al. 2001), and, (3) the spatial process where counts from neighboring points in space are more similar due to auto-correlation in bio-physical factors and their associated animal densities (Legendre 1993). Thus, the count data used to investigate species-habitat relationships are a combined product of these three sources of variation. There are well developed models for each one of these processes individually (McCullagh and Nelder 1989; Legendre 1993; Buckland et al. 2001, 2004; Pollock et al. 2002; Dormann et al. 2007; Thomas et al. 2010) but not in an integrated manner. Recent innovations (Royle 2004; Royle et  al. 2004; Royle and Dorazio 2006; Royle and Dorazio 2008; Joseph et al. 2009; Kéry and Royle 2016) permit these different sources of variation in animal count data to be incorporated into a single hierarchical modeling framework to draw stronger inferences about species-habitat

2.1 Introduction

37

associations. A hierarchical model is an ordered series of models that explicitly builds relationships between the biological process and observation process or other sampling processes. In this modeling framework, information obtained at the level of an individual sampling unit is used to specify models of abundance and detectability, and, to describe variation in abundance and detectability among sampling units. The abundance and detectability parameters are then linked to draw inferences about animal density, as well as factors influencing its spatial variation. Such hierarchical models are increasingly being used to estimate abundance from simple binomial count data (Royle 2004), point transect data (Royle et  al. 2004; Sillett et al. 2012), removal sampling data (Royle and Dorazio 2006; Dorazio et al. 2005), presence-absence data (Royle et  al. 2007), aerial waterfowl survey data (Royle 2008), spatial capture-recapture data (Borchers and Efford 2008; Royle et al. 2009a, 2009b; Royle et al. 2014) and some combinations thereof (e.g., Amundson et al. 2014). These approaches can also explore influences of local habitat factors and management interventions. Tropical forest ungulate densities typically tend to be low because of relatively large body size (body mass > 20 kg) and hence substantially high energetic requirements (Eisenberg 1980; Hudson 1984). Consequently, ungulate count data are typically sparse at the level of an individual sampling unit such as transect line (Buckland et al. 2001; Thomas and Karanth 2002). Imperfect detection further reduces sample sizes because some animals are missed during the observation process. Such sparse data makes it difficult to draw strong inferences about ungulate abundance at finer scales. One approach to resolve this problem is to pool observations across all the sampling units for modeling of detectability, to correct the aggregated total count. However, such pooling of spatially indexed data results in the loss of information on the variation in count data arising from the underlying ecological and spatial processes. For example, Kumar (2000) aggregated count data of chital deer (Axis axis) from seven spatially distinct transect lines, which varied from 1 to 318 animal clusters, to estimate abundance in Ranthambore National Park in India. Thus, the resultant abundance estimate is an adjusted overall count corrected for imperfect detection (Royle and Dorazio 2008). Conversely, methods that effectively investigate the influence of ecological and spatial factors on animal abundance often rely on ‘naïve’ estimators (e.g., Lichstein et al. 2002). They simply ignore the fact that the counts represent only a fraction of the total number of animals available for detection (Royle et al. 2007). Furthermore, this fraction may co-vary with the same ecological and spatial factors that also influence density. Such practices may yield a biased representation of factors that influence animal density. This may even lead to false conclusions about the underlying ecological relationships between animal density and environment variables (Gu and Swihart 2004; Royle et al. 2004). In this chapter, we develop a spatially explicit hierarchical model to estimate tropical forest ungulate densities from line transect data. We first provide a brief overview of the hierarchical modeling approach highlighting its advantages over other potential approaches. We then identify various components that are required for modeling ungulate density using line transect data and describe model

38

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

formulation in detail. We use a Bayesian inferential approach (Gelman et al. 2004; Banerjee et al. 2004; Link and Barker 2010) and implement the hierarchical spatial model in R (R Core Team 2019) using the package NIMBLE (de Valpine et  al. 2017; NIMBLE Development Team 2019a). We demonstrate how a simple and yet flexible structure of the hierarchical model can effectively deal with all the complexities involved in modeling the spatial variation of ungulate density. This chapter also discusses the applicability of the model developed here for a variety of other sampling situations that typically yield multinomial data. We finally illustrate the application of this model by assessing habitat relationships of chital deer Axis axis, based on line transect survey data from the Nagarahole-Bandipur landscape in India.

2.2

Modeling Philosophy

2.2.1 O  verview of Distance Sampling and Line Transect Sampling Distance sampling is a class of methods used to estimate population density of animals (see Buckland et al. 2001, 2004, 2015; Williams et al. 2002 for reviews). Line transect sampling is a specific distance sampling formulation which is particularly useful for assessing the abundance of diurnal tropical forest ungulates (Thomas and Karanth 2002). Under this method, two observers typically survey the area of interest by walking along several spatially replicated trails called ‘transect lines’ to detect, identify and count animals, which occur in ‘clusters’ consisting of one or more individuals. These clusters are geometric artifacts, rather than entire social groups, and identifying them in a predefined manner is a part of the survey protocol (Karanth et al. 2002; Kumar et al. 2017). Upon visually detecting animal clusters from the line, surveyors also record distances and compass bearings from transect to animal clusters detected on either side. Since measurements of distance, compass bearings, cluster size and even identification of species tend to be inaccurate in aural detections (Alldredge et al. 2007), typically only visual detections are included in line transect surveys of animals. In a standard line transect survey, the data consist of a set of radial distances and sighting angles, which are subsequently used to compute perpendicular distances from the transect line to observed animal clusters. These distance data are then used to model detection probability as a function of perpendicular distances (Buckland et al. 2001). The classical derivation of the density estimator in line transect sampling is based on the probability density function of observed perpendicular distances, evaluated at zero (Buckland et al. 2001). This method uses partial likelihood approaches (Burnham et al. 1980; Buckland et al. 2001) to estimate the distance function, and it is assumed that the probability of detecting an animal cluster on the line (i.e., at zero perpendicular distance from the transect line) is 1. The sampling method also provides several options to reduce bias and improve precision of the estimate (Buckland et al. 2001; Buckland et al. 2004). For example, design-based solutions are provided to ensure random placement of transects (Strindberg et al. 2004). Another example is to include

2.2  Modeling Philosophy

39

several covariates (e.g., survey season, cluster size, habitat type etc.) to model sources of heterogeneity in the detection process (Marques et al. 2007). Thus, the conventional distance sampling method using conditional likelihood approaches (Burnham et al. 1980; Buckland et al. 2001) assumes that the sampling process itself is the source of all uncertainties. Hence, the focus is on modeling the distance component of the count data, emphasizing the choice of sampled sites (i.e., transects), which need to be randomly and independently placed with reference to the unknown sampled animal population. This approach avoids the prior specification of a probability model for the distribution of animals, which has an influence on the count statistic obtained (Buckland et al. 2001). While pointing out the advantages of this conditional likelihood (design-based) approach for estimating density, Buckland et al. (2001) note the difficulties involved in developing a fully model-­ based inferential approach. Such an approach involves a full likelihood analysis based on the joint distribution of the observed distances and animal-clusters. Advanced distance sampling methods deal with some of these problems (Buckland et  al. 2004; Hedley and Buckland 2004; Royle et  al. 2004; Miller et  al. 2013; Oedekoven et al. 2014).

2.2.2 Model-Based Inferential Approaches Instead of obtaining a single estimate of overall animal density, ecologists and managers are often interested in the spatial variation of animal density within their area of interest (Brown et al. 1995). Variations in animal densities are likely to be induced systematically by several different environmental factors as well as by stochastic processes (Royle and Dorazio 2006). We may also be interested in predicting animal densities at unsampled locations using data from sampled locations and their relationship to relevant habitat covariates. Conventional distance sampling approaches are inadequate to investigate such responses of animal populations to habitat factors because of their inability to probabilistically model the underlying spatial distribution of animals or variations in abundance within the surveyed regions (Miller et al. 2013). In contrast, the distribution of all possible realizations of values of the sample counts can be described by a stochastic model with or without covariates (Royle and Dorazio 2006; Royle and Dorazio 2008) within a fully model-based inferential approach (Link and Barker 2010). Further, unlike in conventional approaches that treat local density and effects of habitat on them as fixed estimable quantities, the fully model-based approaches can build uncertainty into the estimation of all these model parameters by specifying underlying probability distributions (Ellison 2004). Such approaches represent the fact that only the observed data are fixed and all the model parameters are random outcomes, thus leading to unbiased predictions. However, such complex models involve unconditional likelihood formulations, which are computationally intensive and often intractable. This hindered the widespread implementation of complex models until the advent of fast computers with powerful processors. Coupled with availability of new software tools for modeling and estimation, this situation has dramatically

40

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

changed over the recent years. Consequently, the use of such approaches for investigating ecological problems has been continually increasing.

2.2.3 Hierarchical Models Hierarchical models provide a powerful framework for implementation of fully model-based inferential approaches in ecology (Kéry and Royle 2016). Instead of focusing entirely on either the observation process (disregarding the ecological process) or the ecological process (disregarding the observation and other sampling processes), hierarchical models follow a “conceptual middle ground” to explicitly describe the relationship between observations and biological processes within a single cohesive framework (Royle and Dorazio 2008). Hierarchical models typically have two or more linked constituent models describing the variation in data due to variation in the biological and observation processes. In the following sections, we describe the formulation of a hierarchical distance sampling model.

2.3

Model Development

2.3.1 Motives and Considerations Forest ungulates are diurnal and relatively easy to observe in undisturbed areas during their peak foraging activity periods. During line transect surveys, they are typically detected as clusters consisting of one or more individuals. Each detected cluster is a loose association of individuals observed during a field survey, rather than a social group of individuals. Since it is not practical to record field measurements (sighting distance and compass bearings) to each of the individuals detected during line transect surveys, observations are recorded with respect to each animal cluster detected either through exact distances measured through a ranging instrument or a discrete category of distances. Hence, it is important to recognize the discrete nature of these group-level data while modeling the number of animal clusters per unit area. Individual animal densities can subsequently be derived by multiplying estimated animal cluster density with the estimated cluster size. A host of ecological and management factors influence tropical forest ungulate densities (see Chap. 1 for a detailed description) either over the entire landscape (e.g., forest type) or at fine scales (e.g., rich forage patch). Hence, investigation of ungulate-habitat relationships at different spatial scales, together with predictions of animal density at local scales, will be useful for planning and assessing specific management actions. We considered count data from spatially distinct transect lines that cover the study landscape for modeling ungulate-habitat relationships. These transect lines are placed randomly or systematically over the survey area or purposefully placed depending on the study objectives. Following Levin (1992), we defined each of the transect lines as a “site” and the study area as the “landscape” for investigating

2.3  Model Development

41

influence of covariates on ungulate abundances at two distinct spatial scales. We interchangeably use the terms “site-level” with “local scale” and “landscape-level” with “broad scale”, respectively, depending on the context. Several potential explanatory variables are identified a priori at either or both these spatial scales. Site-level covariate data were collected along each transect, while the landscape-­ level covariate data were collected for each of the 1-km grid squares laid across the study landscape. We note that due to the nature of covariate data collected at multiple spatial scales, there would potentially be a spatial misalignment between transect-level and grid-cell level information. Variation in sampling efforts due to logistic constraints is another potential source of bias. Our primary objective was to use both site-level and landscape-level covariate data to explain variations in ungulate abundance manifested through transect specific (site-level) ungulate count data and predict abundance over unsampled locations. However, we recognized that the ungulate counts were an outcome of a combined process that included observation and spatial processes in addition to the ecological process of interest. Hence, the model should adequately deal with all the potential sources of variations to enhance its predictive ability. All these practical considerations influenced our model formulation.

2.3.2 Model Formulation We use the basic modeling structure described by Royle et al. (2004) to model ungulate abundance as a function of a set of covariates defined a priori. In addition to the perpendicular distance, we used cluster size as a detection-related covariate to describe the observation process. The abundance model consisted of several variations of the Royle et al. (2004) model to investigate effects of ecological factors on ungulate abundance at different spatial scales. This model was adapted to incorporate effects of spatial interactions and to account for variation in sampling efforts. We show here how the model structure is flexible and accommodates additional structures to build a realistic spatial model of animal abundance that can be implemented in R programming language using freely available software packages such as WinBUGS (Spiegelhalter et al. 2007) or NIMBLE (NIMBLE Development Team 2019a).

2.3.3 Model Notations We first introduce the notations that we have used to describe the hierarchical spatial model and then present detailed formulation of the model. Notations Used N = Unknown abundance of animal clusters in the study landscape ni = Unknown abundance of animal clusters in transect i yi = Count of animal clusters detected out of n animal clusters present along transect i xi = Perpendicular distance derived for each of the detected yi animal clusters

42

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

li = Total distance walked on transect i g(x) = Probability of observing an animal cluster at distance x from a transect m = Number of individual animals observed in each animal cluster g(m) = Probability that an animal cluster contains exactly m individuals gs(k) = Probability that an animal cluster is a member of the cluster size category k mk  = Expected number of animals in cluster size category k gi , kj  = Probability that an animal cluster in ith transect belonging to cluster size category k is detected in distance category j πi, kj = Multinomial probability that the ith animal cluster is a member of the cluster size category k, and, is detected in distance category j λi = Expected abundance of animal clusters in transect i sij = Value of transect-level abundance covariate j measured at transect i αj = Regression coefficient (slope) of transect-level abundance covariate j waj = Indicator variable for each of the transect-level covariate effects zk = Expected abundance of animal clusters in grid-cell k ukj = Value of grid-cell level abundance covariate j measured at grid-cell k β0 = Value of regression intercept βj = Regression coefficient (slope) of grid-cell level abundance covariate j bk = Value of latent spatial effects for grid-cell k wj = Indicator variable for each of the grid-cell level covariate effects vik = Proportion of ith transect length in kth grid-cell Zi = Aggregated grid-level abundance of animal clusters in ith transect

2.3.4 Observation Model Distance from transect line and cluster size are the two important parameters that primarily influence the observation process in line transect sampling (Thomas and Karanth 2002), and we assume that both are independent random variables. Thus, the detectability of animal clusters is a function of both these variables. In the sections below, we describe modeling of these components individually and then modeling detection probability as a product of these two variables.

2.3.4.1 Modeling Distance Effects on Detectability Let N represent the unknown abundance of animal clusters in the study landscape and ni the unknown transect-level abundance of animal clusters for transect i. Suppose i (i = 1, 2, …, I) denotes a transect and I is the number of transect lines sampled within the study landscape, then the basic data resulting from a line transect survey is a set of x distances recorded on y animal clusters ‘detected’ out of ni animal clusters present along a set of transect lines. Each transect line is typically walked a number of times in distance sampling surveys; so yi denotes counts of observed animal clusters pooled from all temporal replicates under each transect (Buckland et al. 2001). Note that the distance measured (xi) is often exact and hence a continuous variable, but it can be assigned to discrete distance classes for computational ease or to account for error in the distance measurements. Also, ‘binning’ of

2.3  Model Development

43

Table 2.1  Two-dimensional structure of example line transect count data grouped in distance categories T[1] T[2] T[3] T[4] T[5]

X[1] 3 1 7 15 4

X[2] 2 3 3 4 1

X[3] 1 1 2 2 1

X[4] 0 1 2 0 0

X[5] 0 0 0 0 0

X[6] 0 0 0 0 0

X[7] 0 0 0 1 0

X[8] 0 0 1 0 0

X[9] 0 0 0 0 1

X[10] 0 0 0 0 0

exact distances into a large number of distance classes with a narrow class interval provides a good approximation of the continuous distance function, which is in contrast to many applications where data are collected in very coarse bins. Suppose the observed perpendicular distances to the detected animal clusters along each transect line are divided into Ji categories and Bi is the maximum perpendicular distance at which animal clusters were detected, then: [0, bi1], (bi1, bi2], ……, (bi,Ji - 1, bi,Ji], note that bi, Ji ≡ Bi. Thus, the outcome from a transect survey yij is the observed counts of animal clusters in distance class j at sampled site (transect line) i. The Table 2.1 illustrates line transect data structure. In the above example, there are 10 distance classes (say with a class interval of 15 m) represented by columns X[1]–X[10] and 5 transect lines represented by rows T[1]–T[5]. Each cell represents observed counts of animal clusters in each transect line under each distance class. One of the most widely used detection functions to model the relationship between detection probability and distance observed is the “half-normal” detection function (Buckland et al. 2001). The half-normal detection function uses the perpendicular distances (xi) to estimate detection probability g(xi), given by the expression:



 x2 g  xi   exp   i 2  2

  

The basic assumption in distance sampling is that the detection of an animal cluster located on the line (i.e., at zero perpendicular distance) is certain and hence g(0) = 1. Therefore, the animal clusters that are not located on line, are likely to be missed, and their detection probability g(xi) is a declining function of distance from the transect line. By assuming xi ∼ Uniform(0, Bi) and using the probability density function of observed perpendicular distances, we can estimate the probability of detecting an animal cluster given that it is within Bi, the maximum perpendicular distance observed in the ith transect line.

2.3.4.2 Modeling the Effects of Cluster Size on Detectability In addition to the distance from transect line, the cluster size also influences the detectability of ungulates (Buckland et al. 2001; Royle 2008). Although exact cluster sizes observed were recorded in the field survey, we classified them into discrete categories, because the construction of a large multinomial data matrix (the number

44

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

Table 2.2  Three-dimensional structure of example line transect count data grouped in cluster size categories and distance categories [,, T[4]] gs[1] gs[2] gs[3] gs[4]

X[1] 11 3 1 0

X[2] 2 0 2 0

X[3] 1 0 1 0

X[4] 0 0 0 0

X[5] 0 0 0 0

X[6] 0 0 0 0

X[7] 0 1 0 0

X[8] 0 0 0 0

X[9] 0 0 0 0

X[10] 0 0 0 0

of animals observed in each cluster size in each distance category) would be unwieldy (Royle 2008), especially for gregarious species such as chital, in which observed cluster size can range from 1 to 100 s of individuals (Karanth and Sunquist 1992). Also, observers may underestimate the cluster size particularly at farther distances from the transect line either due to the screening effect of understorey vegetation or measurement errors or a combination of both these factors. We address these challenges by transforming exact cluster size data into grouped data and estimate the expected cluster size by assuming cluster size as a random variable with an underlying probability distribution. The data table now is a three-dimensional matrix that provides observed counts of animal clusters in each distance class under each cluster size category for each transect. We illustrate the data structure (Table 2.2) for the combined detection function using the example of observed counts from one of the transect lines given earlier for the distance function. In the above example, rows gs[1] to gs[4] represent cluster size categories, and 10 distance classes are represented by columns X[1]–X[10], as earlier. Each cell now represents observed counts of animal clusters in each distance class under each cluster size category for transect line T[4]. Suppose the observed number of individuals per animal cluster is divided into K categories: [1, c1], (c1, c2], ……., (cK-1, cK]. Let m denote the number of individuals in an animal cluster. Because all cluster sizes must necessarily be ≥1, we assumed cluster sizes follow a zero-truncated Poisson distribution across all sampled transect lines, with mean μ/(1 −  exp(−μ)). Now, when m ∈ {1, 2, .…}, the probability that a cluster contains exactly m individuals is derived from its probability mass function: g m 

exp      m

m ! 1  exp     

Given this, the probability that an animal cluster is a member of cluster size category k is gs k 

ck

 g m

m  ck 1 1 and the average number of individuals in cluster size category k is

(2.1)

2.3  Model Development

45



mk  1 / gs k

ck

 

m  g m

m  ck 1 1



Note that the expected cluster size, μ/(1 − exp(−μ)), is the average number of individuals in an animal cluster after accounting for the unobserved cluster sizes.

2.3.4.3 Modeling Detection Probability as a Function of Both Distance and Cluster Size To complete the observation model, we now model the detection probability as a function of both distance and cluster size. Recall, yi =  yi ,11 ,yi ,12 ,.,yi , KJ  ′ is a vector of observed counts of animal clusters in each of the KJi combinations of cluster size and distance categories in transect i. Now, the distribution of observed counts ( yi) conditional on transect-level abundance (ni) can be expressed as follows: i

yi ~ Multinomial  ni , i  here,  i =  i ,11 , i ,12 ,., i , KJi ′ is a vector of multinomial probabilities for each combination of cluster size category k and distance category j, and, ni is the unknown number of animal clusters that are available for detection in the ith transect. This implies that πi, kj is a product of two probabilities: the probability that an animal cluster observed in the ith transect belongs to cluster size category k, and, the probability that an animal cluster belonging to kth cluster size category is detected in distance category j. Therefore,  i , kj  gs k gi , kj , where gs k is the probability that a cluster is a member of cluster size category k and gi , kj is the probability that a cluster belonging to cluster size category k is detected in distance class j. Recall, we have computed gs k using Eq. 2.1. To compute gi , kj , we assume





xik  Uniform  0,Bi  where xik  ∈  [0, Bi] is the perpendicular distance of a cluster of size category k detected in transect i with probability



 x2  g  xik   exp   ik2  .  2 k 

We now model the σ parameter in the half-normal detection function such that detectability is an increasing function of cluster size and a decreasing function of distance from the transect line. Since this detection probability is dependent on mk , the average number of individuals in cluster size category k, we assume a log-­linear model for the scale parameter σk as follows: log  k    0  p  mk  1 here, the parameter p is strictly positive to ensure that both σk and g(xik) increase monotonically with average cluster size mk . Based on these assumptions, we compute gi , kj (the probability that a cluster in transect i belonging to size category k is present and detected in distance class j) as below:

46

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

gi , kj

1  Bi

bij

 x2  exp   ik2  dxik  2 k  bi , j 1



We note that this probability is estimated based on the probability density function of observed perpendicular distances from the transect line, described in detail in Buckland et al. (2001).

2.3.5 Abundance (Process) Model Unlike the conditional likelihood of observations that is conventionally used to estimate abundance in distance sampling, we use the approach developed by Royle et al. (2004), which is based on the joint distribution of observations and detection parameters, as our primary interest is in modeling the spatial variation in abundance due to habitat-related covariates. This unconditional likelihood approach also enables establishment of explicit linkages between abundance and covariates that induce variation at both local and landscape-level scales (Royle et al. 2004; Royle and Dorazio 2008). The modeling structure for the ungulate abundance model is described below. Count data are typically discrete and positively valued. They show strong mean-­ variance relationships, which are useful statistical properties (Royle et al. 2002). We used the Poisson distribution to model local abundance and assumed ni ∼ Poisson(λi), where ni is the unknown number of animal clusters available for detection in transect i, and λi is the expected abundance of animal clusters. In the simplest model, one can assume a constant expected abundance over all sampled sites (transects) such that λi  =  λ. This model describing an unvarying abundance across transect lines, together with the observation model (described in the preceding section), forms a basic 2-level hierarchical model, and the estimator of abundance is based on the joint likelihood of observations and detection parameters. This model can be easily expanded to incorporate a set of variables that are expected to induce variations in animal abundance.

2.3.6 Modeling Effects of Site-Level Covariates Tropical forest ungulates (and most other animals) are not uniformly distributed in space at either local or landscape levels. Variation in their densities is likely to be induced by habitat-related covariates that influence densities at both local (transect level) and broad-scale (landscape) levels. We first consider the effect of site-level covariates (e.g., forage quantity) that affect abundance locally at the level of sample sites (transects). Site-specific expected abundance λi can be modeled as a function of a set of si covariates, such that J



log  i    j sij j 1



2.3  Model Development

47

where sij is a site-specific covariate and αj is the corresponding regression coefficient.

2.3.7 Modeling Effects of Landscape-Level Covariates While the above model specifies λi (a set of site-specific expected abundances over i sample sites) as a function of site-specific covariates, our main interest is to estimate ungulate abundance (response variable) over the survey area, and explain its variation as resulting from a set of broad-scale habitat factors. To address this problem, we will first consider the population of animals over the area of interest as an aggregate of the populations drawn from each of the subset areas. If k equal sized grid-cells of area 1 km2 can be super-imposed covering the entire survey region, the total population N can then be thought of as a collection of local populations from each of the k grid-cells (N = N1 + N2 + … + Nk). Typically, a sample site (individual transect line) cuts through several grid-cells of 1 km2 size. Thus, the transect-level unknown abundance ni is itself an aggregate of the unknown abundances at each of the grid-cells covered by transect i. Several factors including physical (e.g., water availability), environmental (e.g., forest vegetation type) and management attributes (e.g., protection effectiveness) vary across space and are thus likely to influence local ungulate abundances. These grid-cell specific abundances can be assumed to follow a probability distribution conditioned on some landscape-level (grid-cell specific) covariates that induce variation across the survey region. We modeled variation in abundance at the grid-cell level as: J

log  zk    0   j ukj

j 1 here, zk is the expected abundance of animals in the kth grid-cell, ukj is the value of the jth landscape covariate measured at the kth grid-cell, β0 is the intercept and βj is the regression coefficient of the jth covariate. Note here that the notation used for the Poisson intensities (zk) for each grid-cell is different from that used for each transect site (λi). The transect-level expected abundance (λi) is therefore an aggregate of the expected abundance at the grid-cells bisected by transect i that gets additionally informed through transect specific covariates.

2.3.8 Modeling Spatial Effects The abundance model described above accounts for the structured variation in count data expected due to local (site-level) and landscape (grid-cell specific) covariates. However, the response variable zk (i.e., grid-cell specific expected animal abundance) is likely to be spatially autocorrelated (Cressie 1993; Haining 2003) because of a suite of both known (e.g., autocorrelated environmental features) and unknown (unobserved or random spatial effects) factors (Legendre 1993). Spatially regressive models effectively deal with spatial autocorrelation (Besag 1974) and are increasingly used by ecologists to specify spatial dependence. Conditional Autoregressive

48

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

models (CAR) are particularly recommended (Cressie 1993) as they provide a useful framework to model the response variable not only as a function of predictor variables but also conditional on the value of the response variable in neighboring locations (Cressie 1993; Haining 2003). Thus, in a CAR model, if a grid-cell k is surrounded by neighbors whose abundance values are high, then the abundance at grid-cell k will be higher than what is expected based on the covariate effect alone, when there is positive spatial autocorrelation. The neighbors are identified by the adjacency of cells in the grid map, and the neighborhood structure is expressed in the form of an m X m matrix of spatial weights with its elements representing the neighborhood relation. Typically, this will be a binary matrix where mij = 1 when j is a neighbor of grid-cell i and 0 when it is not a neighbor. The neighborhood is defined by the user, but is often a rook’s (4 cells sharing a boundary) or queen’s (8 cells touching any part of the grid-cell) neighborhood. Since the immediate neighbors are more likely to have greater influence than the distant ones, weights can be specified such that distant neighbors get a lesser weight than the nearest ones (Cressie 1993). We used an intrinsic Gaussian CAR model to describe spatial variation in local abundance (Banerjee et al. 2004). The grid-cell specific expected abundance parameter is now in the form of J

log  zk   bk   j ukj

j 1 where bk accounts for the ‘unobserved’ spatial random effects. Note the absence of the intercept term (β0) here, as it is absorbed in the spatial random effect and its inclusion will lead to non-identifiability of the parameter (NIMBLE Development Team 2019b). The spatial autoregressive term (bk) is parameterized by an intrinsic Gaussian CAR prior with a conditional mean and variance (NIMBLE Development Team 2019b), which can be expressed as:   bj  j N k p  bk |b k ,  ~ Normal  , mk   .  mk     Here, τ is the precision parameter, which is constant across each point in space. Nk is the set of all neighbors for grid-cell k and mk+ is the total number of neighbors for grid-cell k. As noted above, all the spatial weights were set to one if grid-cells shared a common boundary. Note that a zero-mean constraint in the intrinsic Gaussian CAR prior specification and an additional intercept term is the default option in WinBUGS (Spiegelhalter et al. 2007), while they need to be imposed in NIMBLE if one is interested in modeling the process mean separately (NIMBLE Development Team 2019b). The abundance model now is in the form of a generalized linear mixed model that deals with both intra-cell (fixed effects due to grid-cell specific covariates) and inter-cell (random spatial interaction effects between grid-cells) effects in a combined form (Banerjee et al. 2004).

2.3  Model Development

49

2.3.9 Modeling the Variation in Sampling Efforts Often due to field constraints (weather, terrain, detours etc.), the actual distance walked along each transect is not similar across sites in spatially and temporally replicated surveys. This uneven sampling effort might induce some biases, since longer distances and larger area sampled expose a larger number of animal clusters to detection. Hence, we used the total distance walked in each transect as an offset term in the model of local abundance (λi). The site-specific abundance model is now expressed as: J

log  i   log  li    j sij

where li is the distance walked variable.

j 1



2.3.10 Modeling the Effects of Spatial Misalignment In distance sampling, the geometrical shape of the transect, often referred to as sampler ‘geometry’ (see Buckland et  al. 2001, 2004), is either a straight line, a square, a zig-zag or of some other shape (Strindberg et al. 2004) placed with either random or purposive orientation. The choice of sampler geometry and transect length is influenced by several considerations including biology of the study species, habitat features of the study landscape, logistics and resources available for the study (Kumar et  al. 2017). Thus, the information (both count and site-specific covariate data) gathered in a transect survey is at the level of the sampling unit, i.e., line transect, while the inferential unit for animal abundance is typically a grid-cell. The grid-cell is an areal unit of fine scale spatial resolution at which animal abundance is predicted using a set of grid-cell specific covariates. Because our interest is in using both site-specific and grid-cell specific predictors to explain variation in abundance, it is necessary to account for spatial misalignment between transect-­ level and grid-cell level information to reduce bias and improve accuracy of the abundance estimator (Banerjee et al. 2004, 2014). As noted earlier, each transect passes through several grid-cells resulting in partitioning of the ith transect length amongst the grid-cells bisected by transect i in different proportions. Hence, the areas and the number of animal clusters exposed to sampling will also vary within the grid-cells bisected by the ith transect. This bias induced by spatial misalignment between the transect level and grid-cell level information has to be corrected such that the expected abundance at each transect (λi) is an aggregate of the expected abundance at grid-cells (zk) shared by the ith transect in varying proportions of the transect length. We address this issue by augmenting the abundance model developed in Sect. 2.3.9. The site-specific abundance model is now expressed as: J



log  i   log  li   log  Z i    j sij j 1



50

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

where Z i   vik zk here, vik is the proportion of the ith transect length in the kth grid-cell, zk is the expected abundance at the kth grid-cell and Zi is the aggregated grid-level abundance of the ith transect. Note that only the non-zero entries in the matrix V contribute to the computation of Zi. Also, note that the zk are the Poisson intensities of the spatial abundance process (modeled using vectors of spatial covariates, their corresponding effects and residual spatial process, described in Sect. 2.3.8) and Zi, the spatial abundance effect aggregated for transect i, links grid-cell level information with transect level information after accounting for spatial misalignment. Now, the log intensity model for the transect-level abundance (λi) is a data fusion model that integrates multiple streams of data.

2.4

Bayesian Inference

Even though the hierarchical model formulated above is fairly simple to describe, there is a complex relationship between observations and a whole suite of latent (unobserved) variables including local abundance of animal clusters, the cluster size variable, spatial effects, etc., and it is not straightforward to express the unconditional likelihood. The Bayesian approach permits inference despite these computational complexities using Markov chain Monte Carlo (MCMC) simulations with relevant sampling algorithms (Banerjee et al. 2004; Gelman et al. 2004; Link and Barker 2010). Further, the predictions of random effects of spatially correlated data are a natural product of the Bayesian analysis by MCMC (Gelman et  al. 2004; Banerjee et al. 2004). MCMC can be implemented in statistical packages such as WinBUGS 1.4.3 (Spiegelhalter et  al. 2007) and NIMBLE 0.8.0 (NIMBLE Development Team 2019a). We used these widely available free software packages to implement the hierarchical spatial model and provide the code for fitting the model in R version 3.6.1 (R Core Team 2019) using NIMBLE in Appendix 2.2.

2.5

Bayesian Variable Selection

Most often, one of the primary goals of population studies will be to reliably assess the spatial distribution of animal abundance based on its relationship with a host of environmental, habitat and management factors to aid conservation and management. The development of the hierarchical spatial model described in the preceding sections meets this need, particularly when predictor variables are carefully chosen from a set of potential variables with a priori expectation that these variables influence the response, and the main interest is in drawing inference on the strength of covariate-specific influences. In such a scenario, a full model with all the chosen covariates is fitted to find evidence of the strength of their individual effects based on the posterior distributions of all the parameters.

2.5  Bayesian Variable Selection

51

However, one might also be interested in exploring the chosen set of variables to identify a variable or a subset of variables as more important than the rest. Such an approach may also assist in identifying a subset of models that may explain a larger fraction of the variation in the response variable from amongst a set of all plausible models. Conversely, one can explore if the response variable is primarily affected by a smaller set of predictors. In this section, we show how the hierarchical model developed in the preceding sections is sufficiently flexible to incorporate variable selection methods in a Bayesian inferential context. There are many Bayesian approaches for determining suitable predictor variables and assessing their relative importance in multiple regressions (Bayesian Information Criterion, Schwarz 1978; Deviation Information Criterion, Spiegelhalter et al. 2002; Bayes factor, Link and Barker 2006; Bayesian stochastic search variable selection procedures, George and McCulloch 1993; Kuo and Mallick 1998; see O’Hara and Sillanpää 2009 for a review). We used a simple modeling structure developed by Kuo and Mallick (1998, see also Congdon 2006) to assess the relative importance of predictor variables wherein binary indicator variables are embedded within the regression equation. We specified a set of indicator variables, one for each covariate effect (both for site-specific and grid-cell specific covariates), and assumed them to be mutually independent. The spatial process model for grid-cell specific abundances is now expressed as: J

log  zk   bk  w j  j ukj

j 1 and the transect-specific abundance model is in the form of J

log  i   log  li   log  Z i   wa j j sij

j 1 here, wj and waj are the indicator variables for each of the grid-cell specific and site-­ specific covariate effects, respectively. These indicator variables are Bernoulli random variables that assume the value of either 1 or 0, and a regressor is included in the model set only when its indicator variable takes the value of 1. We assigned a Bernoulli distribution as the prior for these indicator variables with a 0.5 success probability such that each abundance predictor had an equal probability of inclusion or otherwise in the regression. The posterior mean of the indicator variables represents the inclusion probability of each regressor in the model, which can be used to rank the relative importance of predictors, as well as identify a subset of most useful predictors of abundance. This modeling approach also allows estimation of the marginal posterior probability of all plausible models with different combinations of predictors and identifies an optimal model that has the best support. Thus, this modeling approach essentially serves the purposes of a Bayesian model selection tool to identify an optimal model or a subset of models that explain a large part of the abundance variation patterns. Note that the marginal posterior probability of the optimal model refers to the frequency with which a model containing a predictor

52

2  Development of Hierarchical Spatial Models for Assessing Ungulate Abundance…

variable or a subset of variables appears most frequently amongst a set of all plausible models. On the other hand, the posterior inclusion probability of a predictor variable refers to the expected value of the indicator variable with which a predictor variable appears “in” any of the plausible models.

2.6

 xample of Application of the Hierarchical E Spatial Model

2.6.1 Line Transect Sampling Data on Chital Populations We illustrate the application of the hierarchical spatial model using the count data of chital deer (Axis axis) from line transect surveys conducted during 2005–2006 in the Nagarahole-Bandipur study landscape. Details of the study landscape are described in Chap. 1. The goal was to estimate the density of chital at 1-km2 fine resolution in the study area using counts of chital clusters obtained from a set of 77 transects each of length 3.2 km. Each transect was walked 6–7 times within a short period of